Stable Baselines

Stable Baselines Documentation
Release 2.10.2
Stable Baselines Contributors
Aug 25, 2021

User Guide
1 Main differences with OpenAI Baselines 3

1.1 Installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Getting Started . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Reinforcement Learning Tips and Tricks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Reinforcement Learning Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.5 RL Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.6 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 Vectorized Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.8 Using Custom Environments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.9 Custom Policy Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.10 Callbacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
1.11 Tensorboard Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
1.12 RL Baselines Zoo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.13 Pre-Training (Behavior Cloning) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.14 Dealing with NaNs and infs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
1.15 On saving and loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
1.16 Exporting models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
1.17 Base RL Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
1.18 Policy Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.19 A2C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
1.20 ACER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
1.21 ACKTR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
1.22 DDPG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
1.23 DQN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
1.24 GAIL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
1.25 HER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
1.26 PPO1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
1.27 PPO2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
1.28 SAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
1.29 TD3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
1.30 TRPO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
1.31 Probability Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
1.32 Tensorflow Utils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
1.33 Command Utils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
1.34 Schedules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
1.35 Evaluation Helper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
i
1.36 Gym Environment Checker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
1.37 Monitor Wrapper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
1.38 Changelog . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
1.39 Projects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
1.40 Plotting Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
2 Citing Stable Baselines 215
3 Contributing 217
4 Indices and tables 219
Python Module Index 221
Index 223
ii
Stable Baselines Documentation, Release 2.10.2
Stable Baselines is a set of improved implementations of Reinforcement Learning (RL) algorithms based on OpenAI
Baselines.
Warning: This package is in maintenance mode, please use Stable-Baselines3 (SB3) for an up-to-date version.
You can find a migration guide in SB3 documentation.
Github repository: https://github.com/hill-a/stable-baselines

RL Baselines Zoo (collection of pre-trained agents): https://github.com/araffin/rl-baselines-zoo
RL Baselines zoo also offers a simple interface to train, evaluate agents and do hyperparameter tuning.
You can read a detailed presentation of Stable Baselines in the Medium article: link
User Guide 1
2 User Guide
CHAPTER 1
Main differences with OpenAI Baselines
This toolset is a fork of OpenAI Baselines, with a major structural refactoring, and code cleanups:
• Unified structure for all algorithms
• PEP8 compliant (unified code style)
• Documented functions and classes
• More tests & more code coverage
• Additional algorithms: SAC and TD3 (+ HER support for DQN, DDPG, SAC and TD3)
1.1 Installation
1.1.1 Prerequisites
Baselines requires python3 (>=3.5) with the development headers. You’ll also need system packages CMake, Open-
MPI and zlib. Those can be installed as follows
Note: Stable-Baselines supports Tensorflow versions from 1.8.0 to 1.15.0, and does not work on Tensorflow versions
2.0.0 and above. PyTorch support is done in Stable-Baselines3
Ubuntu
sudo apt-get update && sudo apt-get install cmake libopenmpi-dev python3-dev zlib1g-
˓→dev
3
Mac OS X
Installation of system packages on Mac requires Homebrew. With Homebrew installed, run the following:
brew install cmake openmpi
Windows 10
We recommend using Anaconda for Windows users for easier installation of Python packages and required libraries.
You need an environment with Python version 3.5 or above.
For a quick start you can move straight to installing Stable-Baselines in the next step (without MPI). This supports
most but not all algorithms.
To support all algorithms, Install MPI for Windows (you need to download and install msmpisetup.exe) and
follow the instructions on how to install Stable-Baselines with MPI support in following section.
Note: Trying to create Atari environments may result to vague errors related to missing DLL files and modules. This
is an issue with atari-py package. See this discussion for more information.
Stable Release
To install with support for all algorithms, including those depending on OpenMPI, execute:
pip install stable-baselines[mpi]
GAIL, DDPG, TRPO, and PPO1 parallelize training using OpenMPI. OpenMPI has had weird interactions with Ten-
sorflow in the past (see Issue #430) and so if you do not intend to use these algorithms we recommend installing
without OpenMPI. To do this, execute:
pip install stable-baselines
If you have already installed with MPI support, you can disable MPI by uninstalling mpi4py with pip uninstall
mpi4py.
Note: Unless you are using the bleeding-edge version, you need to install the correct Tensorflow version manually.
See Issue #849
1.1.2 Bleeding-edge version
To install the latest master version:
pip install git+https://github.com/hill-a/stable-baselines
1.1.3 Development version
To contribute to Stable-Baselines, with support for running tests and building the documentation.
4 Chapter 1. Main differences with OpenAI Baselines

git clone https://github.com/hill-a/stable-baselines && cd stable-baselines

pip install -e .[docs,tests,mpi]
1.1.4 Using Docker Images
If you are looking for docker images with stable-baselines already installed in it, we recommend using images from
RL Baselines Zoo.
Otherwise, the following images contained all the dependencies for stable-baselines but not the stable-baselines pack-
age itself. They are made for development.
Use Built Images
GPU image (requires nvidia-docker):
docker pull stablebaselines/stable-baselines
CPU only:
docker pull stablebaselines/stable-baselines-cpu
Build the Docker Images
Build GPU image (with nvidia-docker):
make docker-gpu
Build CPU image:
make docker-cpu
Note: if you are using a proxy, you need to pass extra params during build and do some tweaks:
--network=host --build-arg HTTP_PROXY=http://your.proxy.fr:8080/ --build-arg http_

˓→proxy=http://your.proxy.fr:8080/ --build-arg HTTPS_PROXY=https://your.proxy.fr:8080/
˓→ --build-arg https_proxy=https://your.proxy.fr:8080/
Run the images (CPU/GPU)
Run the nvidia-docker GPU image
docker run -it --runtime=nvidia --rm --network host --ipc=host --name test --mount
˓→src="$(pwd)",target=/root/code/stable-baselines,type=bind stablebaselines/stable-
˓→baselines bash -c 'cd /root/code/stable-baselines/ && pytest tests/'
Or, with the shell file:
./scripts/run_docker_gpu.sh pytest tests/
Run the docker CPU image
1.1. Installation 5
docker run -it --rm --network host --ipc=host --name test --mount src="$(pwd)",
˓→target=/root/code/stable-baselines,type=bind stablebaselines/stable-baselines-cpu
˓→bash -c 'cd /root/code/stable-baselines/ && pytest tests/'
Or, with the shell file:
./scripts/run_docker_cpu.sh pytest tests/
Explanation of the docker command:

• docker run -it create an instance of an image (=container), and run it interactively (so ctrl+c will work)
• --rm option means to remove the container once it exits/stops (otherwise, you will have to use docker rm)
• --network host don’t use network isolation, this allow to use tensorboard/visdom on host machine
• --ipc=host Use the host system’s IPC namespace. IPC (POSIX/SysV IPC) namespace provides separation
of named shared memory segments, semaphores and message queues.
• --name test give explicitly the name test to the container, otherwise it will be assigned a random name
• --mount src=... give access of the local directory (pwd command) to the container (it will be map to
/root/code/stable-baselines), so all the logs created in the container in this folder will be kept
• bash -c '...' Run command inside the docker image, here run the tests (pytest tests/)
1.2 Getting Started
Most of the library tries to follow a sklearn-like syntax for the Reinforcement Learning algorithms.
Here is a quick example of how to train and run PPO2 on a cartpole environment:
import gym
from stable_baselines.common.policies import MlpPolicy

from stable_baselines.common.vec_env import DummyVecEnv
from stable_baselines import PPO2
env = gym.make('CartPole-v1')
# Optional: PPO2 requires a vectorized environment to run
# the env is now wrapped automatically when passing it to the constructor
# env = DummyVecEnv([lambda: env])
model = PPO2(MlpPolicy, env, verbose=1)

model.learn(total_timesteps=10000)
obs = env.reset()
for i in range(1000):
action, _states = model.predict(obs)
obs, rewards, dones, info = env.step(action)
env.render()
Or just train a model with a one liner if the environment is registered in Gym and if the policy is registered:
model = PPO2('MlpPolicy', 'CartPole-v1').learn(10000)

Fig. 1: Define and train a RL agent in one line of code!
1.3 Reinforcement Learning Tips and Tricks
The aim of this section is to help you doing reinforcement learning experiments. It covers general advice about RL
(where to start, which algorithm to choose, how to evaluate an algorithm, . . . ), as well as tips and tricks when using a
custom environment or implementing an RL algorithm.
1.3.1 General advice when using Reinforcement Learning
TL;DR
1. Read about RL and Stable Baselines

2. Do quantitative experiments and hyperparameter tuning if needed
3. Evaluate the performance using a separate test environment
4. For better performance, increase the training budget
Like any other subject, if you want to work with RL, you should first read about it (we have a dedicated resource page to
get you started) to understand what you are using. We also recommend you read Stable Baselines (SB) documentation
and do the tutorial. It covers basic usage and guide you towards more advanced concepts of the library (e.g. callbacks
and wrappers).
Reinforcement Learning differs from other machine learning methods in several ways. The data used to train the agent
is collected through interactions with the environment by the agent itself (compared to supervised learning where you
have a fixed dataset for instance). This dependence can lead to vicious circle: if the agent collects poor quality data
(e.g., trajectories with no rewards), then it will not improve and continue to amass bad trajectories.
This factor, among others, explains that results in RL may vary from one run to another (i.e., when only the seed of the
pseudo-random generator changes). For this reason, you should always do several runs to have quantitative results.
Good results in RL are generally dependent on finding appropriate hyperparameters. Recent algorithms (PPO, SAC,
TD3) normally require little hyperparameter tuning, however, don’t expect the default ones to work on any environ-
ment.
Therefore, we highly recommend you to take a look at the RL zoo (or the original papers) for tuned hyperparameters.
A best practice when you apply RL to a new problem is to do automatic hyperparameter optimization. Again, this is
included in the RL zoo.
When applying RL to a custom problem, you should always normalize the input to the agent (e.g. using VecNormalize
for PPO2/A2C) and look at common preprocessing done on other environments (e.g. for Atari, frame-stack, . . . ).
Please refer to Tips and Tricks when creating a custom environment paragraph below for more advice related to
custom environments.
Current Limitations of RL
You have to be aware of the current limitations of reinforcement learning.

Model-free RL algorithms (i.e. all the algorithms implemented in SB) are usually sample inefficient. They require a
lot of samples (sometimes millions of interactions) to learn something useful. That’s why most of the successes in
RL were achieved on games or in simulation only. For instance, in this work by ETH Zurich, the ANYmal robot was
trained in simulation only, and then tested in the real world.
1.3. Reinforcement Learning Tips and Tricks 7

As a general advice, to obtain better performances, you should augment the budget of the agent (number of training
timesteps).
In order to achieve the desired behavior, expert knowledge is often required to design an adequate reward function.
This reward engineering (or RewArt as coined by Freek Stulp), necessitates several iterations. As a good example
of reward shaping, you can take a look at Deep Mimic paper which combines imitation learning and reinforcement
learning to do acrobatic moves.
One last limitation of RL is the instability of training. That is to say, you can observe during training a huge drop in
performance. This behavior is particularly present in DDPG, that’s why its extension TD3 tries to tackle that issue.
Other method, like TRPO or PPO make use of a trust region to minimize that problem by avoiding too large update.
How to evaluate an RL algorithm?
Because most algorithms use exploration noise during training, you need a separate test environment to evaluate the
performance of your agent at a given time. It is recommended to periodically evaluate your agent for n test episodes
(n is usually between 5 and 20) and average the reward per episode to have a good estimate.
As some policy are stochastic by default (e.g. A2C or PPO), you should also try to set deterministic=True when
calling the .predict() method, this frequently leads to better performance. Looking at the training curve (episode
reward function of the timesteps) is a good proxy but underestimates the agent true performance.
Note: We provide an EvalCallback for doing such evaluation. You can read more about it in the Callbacks
section.
We suggest you reading Deep Reinforcement Learning that Matters for a good discussion about RL evaluation.
You can also take a look at this blog post and this issue by Cédric Colas.
1.3.2 Which algorithm should I use?
There is no silver bullet in RL, depending on your needs and problem, you may choose one or the other. The first
distinction comes from your action space, i.e., do you have discrete (e.g. LEFT, RIGHT, . . . ) or continuous actions
(ex: go to a certain speed)?
Some algorithms are only tailored for one or the other domain: DQN only supports discrete actions, where SAC is
restricted to continuous actions.
The second difference that will help you choose is whether you can parallelize your training or not, and how you can
do it (with or without MPI?). If what matters is the wall clock training time, then you should lean towards A2C and its
derivatives (PPO, ACER, ACKTR, . . . ). Take a look at the Vectorized Environments to learn more about training with
multiple workers.
To sum it up:
Discrete Actions
Note: This covers Discrete, MultiDiscrete, Binary and MultiBinary spaces

Discrete Actions - Single Process
DQN with extensions (double DQN, prioritized replay, . . . ) and ACER are the recommended algorithms. DQN is
usually slower to train (regarding wall clock time) but is the most sample efficient (because of its replay buffer).
Discrete Actions - Multiprocessed
You should give a try to PPO2, A2C and its successors (ACKTR, ACER).
If you can multiprocess the training using MPI, then you should checkout PPO1 and TRPO.
Continuous Actions
Continuous Actions - Single Process
Current State Of The Art (SOTA) algorithms are SAC and TD3. Please use the hyperparameters in the RL zoo for best
results.
Continuous Actions - Multiprocessed
Take a look at PPO2, TRPO or A2C. Again, don’t forget to take the hyperparameters from the RL zoo for continuous
actions problems (cf Bullet envs).
Note: Normalization is critical for those algorithms
If you can use MPI, then you can choose between PPO1, TRPO and DDPG.
Goal Environment
If your environment follows the GoalEnv interface (cf HER), then you should use HER + (SAC/TD3/DDPG/DQN)
depending on the action space.
Note: The number of workers is an important hyperparameters for experiments with HER. Currently, only
HER+DDPG supports multiprocessing using MPI.
1.3.3 Tips and Tricks when creating a custom environment
If you want to learn about how to create a custom environment, we recommend you read this page. We also provide a
colab notebook for a concrete example of creating a custom gym environment.
Some basic advice:
• always normalize your observation space when you can, i.e., when you know the boundaries
• normalize your action space and make it symmetric when continuous (cf potential issue below) A good practice
is to rescale your actions to lie in [-1, 1]. This does not limit you as you can easily rescale the action inside the
environment
• start with shaped reward (i.e. informative reward) and simplified version of your problem
1.3. Reinforcement Learning Tips and Tricks 9

• debug with random actions to check that your environment works and follows the gym interface:
We provide a helper to check that your environment runs without error:
from stable_baselines.common.env_checker import check_env
env = CustomEnv(arg1, ...)

# It will check your custom environment and output additional warnings if needed
check_env(env)
If you want to quickly try a random agent on your environment, you can also do:
env = YourEnv()
obs = env.reset()
n_steps = 10
for _ in range(n_steps):
# Random action
action = env.action_space.sample()
obs, reward, done, info = env.step(action)
Why should I normalize the action space?

Most reinforcement learning algorithms rely on a Gaussian distribution (initially centered at 0 with std 1) for continu-
ous actions. So, if you forget to normalize the action space when using a custom environment, this can harm learning
and be difficult to debug (cf attached image and issue #473).
Another consequence of using a Gaussian is that the action range is not bounded. That’s why clipping is usually used
as a bandage to stay in a valid interval. A better solution would be to use a squashing function (cf SAC) or a Beta
distribution (cf issue #112).

Note: This statement is not true for DDPG or TD3 because they don’t rely on any probability distribution.
1.3.4 Tips and Tricks when implementing an RL algorithm
When you try to reproduce a RL paper by implementing the algorithm, the nuts and bolts of RL research by John
Schulman are quite useful (video).
We recommend following those steps to have a working RL algorithm:
1. Read the original paper several times
2. Read existing implementations (if available)
3. Try to have some “sign of life” on toy problems
4. Validate the implementation by making it run on harder and harder envs (you can compare results against the RL zoo)
You usually need to run hyperparameter optimization for that step.
You need to be particularly careful on the shape of the different objects you are manipulating (a broadcast mistake will
fail silently cf issue #75) and when to stop the gradient propagation.
A personal pick (by @araffin) for environments with gradual difficulty in RL with continuous actions:
1. Pendulum (easy to solve)
2. HalfCheetahBullet (medium difficulty with local minima and shaped reward)
3. BipedalWalkerHardcore (if it works on that one, then you can have a cookie)
in RL with discrete actions:
1. CartPole-v1 (easy to be better than random agent, harder to achieve maximal performance)
2. LunarLander
3. Pong (one of the easiest Atari game)
4. other Atari games (e.g. Breakout)
1.4 Reinforcement Learning Resources
Stable-Baselines assumes that you already understand the basic concepts of Reinforcement Learning (RL).
However, if you want to learn about RL, there are several good resources to get started:
• OpenAI Spinning Up
• David Silver’s course
• Lilian Weng’s blog
• Berkeley’s Deep RL Bootcamp
• Berkeley’s Deep Reinforcement Learning course
• More resources
1.4. Reinforcement Learning Resources 11

1.5 RL Algorithms
This table displays the rl algorithms that are implemented in the stable baselines project, along with some useful
characteristics: support for recurrent policies, discrete/continuous actions, multiprocessing.
Name Refactored1 Recurrent Box Discrete Multi Processing

A2C X X X X X
4
ACER X X X X
ACKTR X X X X X
DDPG X X X3
DQN X X
HER X X X
GAIL2 X X X X X3
PPO1 X X X X3
PPO2 X X X X X
SAC X X
TD3 X X
TRPO X X X X3
Note: Non-array spaces such as Dict or Tuple are not currently supported by any algorithm, except HER for dict
when working with gym.GoalEnv
Actions gym.spaces:
• Box: A N-dimensional box that contains every point in the action space.
• Discrete: A list of possible actions, where each timestep only one of the actions can be used.
• MultiDiscrete: A list of possible actions, where each timestep only one action of each discrete set can be
used.
• MultiBinary: A list of possible actions, where each timestep any of the actions can be used in any combina-
tion.
Note: Some logging values (like ep_rewmean, eplenmean) are only available when using a Monitor wrapper
See Issue #339 for more info.
1.5.1 Reproducibility
Completely reproducible results are not guaranteed across Tensorflow releases or different platforms. Furthermore,
results need not be reproducible between CPU and GPU executions, even when using identical seeds.
In order to make computations deterministic on CPU, on your specific problem on one specific platform, you need to
pass a seed argument at the creation of a model and set n_cpu_tf_sess=1 (number of cpu for Tensorflow session). If
you pass an environment to the model using set_env(), then you also need to seed the environment first.
1 Whether or not the algorithm has be refactored to fit the BaseRLModel class.
4 TODO, in project scope.
3 Multi Processing with MPI.
2 Only implemented for TRPO.

Note: Because of the current limits of Tensorflow 1.x, we cannot ensure reproducible results on the GPU yet. This
issue is solved in Stable-Baselines3 “PyTorch edition”
Note: TD3 sometimes fail to have reproducible results for obscure reasons, even when following the previous steps
(cf PR #492). If you find the reason then please open an issue ;)
Credit: part of the Reproducibility section comes from PyTorch Documentation
1.6 Examples
1.6.1 Try it online with Colab Notebooks!
All the following examples can be executed online using Google colab notebooks:
• Full Tutorial
• All Notebooks
• Getting Started
• Training, Saving, Loading
• Multiprocessing
• Monitor Training and Plotting
• Atari Games
• Breakout (trained agent included)
• Hindsight Experience Replay
• RL Baselines zoo
1.6.2 Basic Usage: Training, Saving, Loading
In the following example, we will train, save and load a DQN model on the Lunar Lander environment.
Fig. 2: Lunar Lander Environment
Note: LunarLander requires the python package box2d. You can install it using apt install swig and then
pip install box2d box2d-kengz
Note: load function re-creates model from scratch on each call, which can be slow. If you need to e.g. evaluate
same model with multiple different sets of parameters, consider using load_parameters instead.
1.6. Examples 13
import gym
from stable_baselines import DQN

from stable_baselines.common.evaluation import evaluate_policy
# Create environment
env = gym.make('LunarLander-v2')
# Instantiate the agent

model = DQN('MlpPolicy', env, learning_rate=1e-3, prioritized_replay=True, verbose=1)
# Train the agent
model.learn(total_timesteps=int(2e5))
# Save the agent
model.save("dqn_lunar")
del model # delete trained model to demonstrate loading
# Load the trained agent

model = DQN.load("dqn_lunar")
# Evaluate the agent

mean_reward, std_reward = evaluate_policy(model, model.get_env(), n_eval_episodes=10)
# Enjoy trained agent

obs = env.reset()
env.render()
1.6.3 Multiprocessing: Unleashing the Power of Vectorized Environments
Fig. 3: CartPole Environment
import gym
import numpy as np

from stable_baselines.common.vec_env import SubprocVecEnv
from stable_baselines.common import set_global_seeds, make_vec_env
from stable_baselines import ACKTR
def make_env(env_id, rank, seed=0):

"""
Utility function for multiprocessed env.
:param env_id: (str) the environment ID

:param num_env: (int) the number of environments you wish to have in subprocesses
:param seed: (int) the inital seed for RNG
(continues on next page)

(continued from previous page)

:param rank: (int) index of the subprocess
"""
def _init():
env = gym.make(env_id)
env.seed(seed + rank)
return env
set_global_seeds(seed)
return _init
if __name__ == '__main__':
env_id = "CartPole-v1"
num_cpu = 4 # Number of processes to use
# Create the vectorized environment
env = SubprocVecEnv([make_env(env_id, i) for i in range(num_cpu)])
# Stable Baselines provides you with make_vec_env() helper

# which does exactly the previous steps for you:
# env = make_vec_env(env_id, n_envs=num_cpu, seed=0)
model = ACKTR(MlpPolicy, env, verbose=1)

obs = env.reset()
for _ in range(1000):
env.render()
1.6.4 Using Callback: Monitoring Training
Note: We recommend reading the Callback section
You can define a custom callback function that will be called inside the agent. This could be useful when you want
to monitor training, for instance display live learning curves in Tensorboard (or in Visdom) or save the best agent. If
your callback returns False, training is aborted early.
import os
import gym
import numpy as np
import matplotlib.pyplot as plt
from stable_baselines import DDPG

from stable_baselines.ddpg.policies import LnMlpPolicy
from stable_baselines import results_plotter
from stable_baselines.bench import Monitor
from stable_baselines.results_plotter import load_results, ts2xy
from stable_baselines.common.noise import AdaptiveParamNoiseSpec
1.6. Examples 15
Fig. 4: Learning curve of DDPG on LunarLanderContinuous environment


from stable_baselines.common.callbacks import BaseCallback
class SaveOnBestTrainingRewardCallback(BaseCallback):
"""
Callback for saving a model (the check is done every ``check_freq`` steps)
based on the training reward (in practice, we recommend using `ÈvalCallback``).
:param check_freq: (int)

:param log_dir: (str) Path to the folder where the model will be saved.
It must contains the file created by the ``Monitor`` wrapper.
:param verbose: (int)
"""
def __init__(self, check_freq: int, log_dir: str, verbose=1):
super(SaveOnBestTrainingRewardCallback, self).__init__(verbose)
self.check_freq = check_freq
self.log_dir = log_dir
self.save_path = os.path.join(log_dir, 'best_model')
self.best_mean_reward = -np.inf
def _init_callback(self) -> None:

# Create folder if needed
if self.save_path is not None:
os.makedirs(self.save_path, exist_ok=True)
def _on_step(self) -> bool:

if self.n_calls % self.check_freq == 0:
# Retrieve training reward

x, y = ts2xy(load_results(self.log_dir), 'timesteps')
if len(x) > 0:
# Mean training reward over the last 100 episodes
mean_reward = np.mean(y[-100:])
if self.verbose > 0:
print("Num timesteps: {}".format(self.num_timesteps))
print("Best mean reward: {:.2f} - Last mean reward per episode: {:.2f}
˓→".format(self.best_mean_reward, mean_reward))
# New best model, you could save the agent here

if mean_reward > self.best_mean_reward:
self.best_mean_reward = mean_reward
# Example for saving best model
if self.verbose > 0:
print("Saving new best model to {}".format(self.save_path))
self.model.save(self.save_path)
return True
# Create log dir

log_dir = "tmp/"
os.makedirs(log_dir, exist_ok=True)
# Create and wrap the environment

env = gym.make('LunarLanderContinuous-v2')
env = Monitor(env, log_dir)
# Add some param noise for exploration

1.6. Examples 17

param_noise = AdaptiveParamNoiseSpec(initial_stddev=0.1, desired_action_stddev=0.1)
# Because we use parameter noise, we should use a MlpPolicy with layer normalization
model = DDPG(LnMlpPolicy, env, param_noise=param_noise, verbose=0)
# Create the callback: check every 1000 steps
callback = SaveOnBestTrainingRewardCallback(check_freq=1000, log_dir=log_dir)
# Train the agent
time_steps = 1e5
model.learn(total_timesteps=int(time_steps), callback=callback)
results_plotter.plot_results([log_dir], time_steps, results_plotter.X_TIMESTEPS,

˓→"DDPG LunarLander")
plt.show()
1.6.5 Atari Games
Fig. 5: Trained A2C agent on Breakout
Fig. 6: Pong Environment
Training a RL agent on Atari games is straightforward thanks to make_atari_env helper function. It will do all
the preprocessing and multiprocessing for you.
from stable_baselines.common.cmd_util import make_atari_env

from stable_baselines.common.vec_env import VecFrameStack
from stable_baselines import ACER
# There already exists an environment generator

# that will make and wrap atari environments correctly.
# Here we are also multiprocessing training (num_env=4 => 4 processes)
env = make_atari_env('PongNoFrameskip-v4', num_env=4, seed=0)
# Frame-stacking with 4 frames
env = VecFrameStack(env, n_stack=4)
model = ACER('CnnPolicy', env, verbose=1)

obs = env.reset()
while True:
env.render()
1.6.6 PyBullet: Normalizing input features
Normalizing input features may be essential to successful training of an RL agent (by default, images are scaled but
not other types of input), for instance when training on PyBullet environments. For that, a wrapper exists and will

compute a running average and standard deviation of input features (it can do the same for rewards).
Note: you need to install pybullet with pip install pybullet
import os
import gym
import pybullet_envs
from stable_baselines.common.vec_env import DummyVecEnv, VecNormalize

env = DummyVecEnv([lambda: gym.make("HalfCheetahBulletEnv-v0")])

# Automatically normalize the input features and reward
env = VecNormalize(env, norm_obs=True, norm_reward=True,
clip_obs=10.)
model = PPO2('MlpPolicy', env)

# Don't forget to save the VecNormalize statistics when saving the agent
log_dir = "/tmp/"
model.save(log_dir + "ppo_halfcheetah")
stats_path = os.path.join(log_dir, "vec_normalize.pkl")
env.save(stats_path)
# To demonstrate loading
del model, env
# Load the agent

model = PPO2.load(log_dir + "ppo_halfcheetah")
# Load the saved statistics

env = DummyVecEnv([lambda: gym.make("HalfCheetahBulletEnv-v0")])
env = VecNormalize.load(stats_path, env)
# do not update them at test time
env.training = False
# reward normalization is not needed at test time
env.norm_reward = False
1.6.7 Custom Policy Network
Stable baselines provides default policy networks for images (CNNPolicies) and other type of inputs (MlpPolicies).
However, you can also easily define a custom architecture for the policy network (see custom policy section):
import gym
from stable_baselines.common.policies import FeedForwardPolicy

from stable_baselines import A2C
# Custom MLP policy of three layers of size 128 each

class CustomPolicy(FeedForwardPolicy):
def __init__(self, *args, **kwargs):
1.6. Examples 19

super(CustomPolicy, self).__init__(*args, **kwargs,
net_arch=[dict(pi=[128, 128, 128], vf=[128,
˓→ 128, 128])],
feature_extraction="mlp")
model = A2C(CustomPolicy, 'LunarLander-v2', verbose=1)

# Train the agent
1.6.8 Accessing and modifying model parameters
You can access model’s parameters via load_parameters and get_parameters functions, which use dictio-
naries that map variable names to NumPy arrays.
These functions are useful when you need to e.g. evaluate large set of models with same network structure, visualize
different layers of the network or modify parameters manually.
You can access original Tensorflow Variables with function get_parameter_list.
Following example demonstrates reading parameters, modifying some of them and loading them to model by imple-
menting evolution strategy for solving CartPole-v1 environment. The initial guess for parameters is obtained by
running A2C policy gradient updates on the model.
import gym
import numpy as np
def mutate(params):
"""Mutate parameters by adding normal noise to them"""
return dict((name, param + np.random.normal(size=param.shape))
for name, param in params.items())
def evaluate(env, model):

"""Return mean fitness (sum of episodic rewards) for given model"""
episode_rewards = []
for _ in range(10):
reward_sum = 0
done = False
obs = env.reset()
while not done:
reward_sum += reward
episode_rewards.append(reward_sum)
return np.mean(episode_rewards)
# Create env
# Create policy with a small network
model = A2C('MlpPolicy', env, ent_coef=0.0, learning_rate=0.1,
policy_kwargs={'net_arch': [8, ]})
# Use traditional actor-critic policy gradient updates to

# find good initial parameters

# Get the parameters as the starting point for ES

mean_params = model.get_parameters()
# Include only variables with "/pi/" (policy) or "/shared" (shared layers)

# in their name: Only these ones affect the action.
mean_params = dict((key, value) for key, value in mean_params.items()
if ("/pi/" in key or "/shared" in key))
for iteration in range(10):

# Create population of candidates and evaluate them
population = []
for population_i in range(100):
candidate = mutate(mean_params)
# Load new policy parameters to agent.
# Tell function that it should only update parameters
# we give it (policy parameters)
model.load_parameters(candidate, exact_match=False)
fitness = evaluate(env, model)
population.append((candidate, fitness))
# Take top 10% and use average over their parameters as next mean parameter
top_candidates = sorted(population, key=lambda x: x[1], reverse=True)[:10]
mean_params = dict(
(name, np.stack([top_candidate[0][name] for top_candidate in top_candidates]).
˓→mean(0))
for name in mean_params.keys()

)
mean_fitness = sum(top_candidate[1] for top_candidate in top_candidates) / 10.0
print("Iteration {:<3} Mean top fitness: {:.2f}".format(iteration, mean_fitness))
1.6.9 Recurrent Policies
This example demonstrate how to train a recurrent policy and how to test it properly.
Warning: One current limitation of recurrent policies is that you must test them with the same number of
environments they have been trained on.
# For recurrent policies, with PPO2, the number of environments run in parallel
# should be a multiple of nminibatches.
model = PPO2('MlpLstmPolicy', 'CartPole-v1', nminibatches=1, verbose=1)
model.learn(50000)
# Retrieve the env

env = model.get_env()
obs = env.reset()
# Passing state=None to the predict function means
# it is the initial state
state = None
# When using VecEnv, done is a vector
done = [False for _ in range(env.num_envs)]
1.6. Examples 21

# We need to pass the previous state and a mask for recurrent policies
# to reset lstm state when a new episode begin
action, state = model.predict(obs, state=state, mask=done)
obs, reward , done, _ = env.step(action)
# Note: with VecEnv, env.reset() is automatically called
# Show the env

env.render()
1.6.10 Hindsight Experience Replay (HER)
For this example, we are using Highway-Env by @eleurent.
Fig. 7: The highway-parking-v0 environment.
The parking env is a goal-conditioned continuous control task, in which the vehicle must park in a given space with
the appropriate heading.
Note: the hyperparameters in the following example were optimized for that environment.
import gym
import highway_env
import numpy as np
from stable_baselines import HER, SAC, DDPG, TD3

from stable_baselines.ddpg import NormalActionNoise
env = gym.make("parking-v0")
# Create 4 artificial transitions per real transition

n_sampled_goal = 4
# SAC hyperparams:
model = HER('MlpPolicy', env, SAC, n_sampled_goal=n_sampled_goal,
goal_selection_strategy='future',
verbose=1, buffer_size=int(1e6),
learning_rate=1e-3,
gamma=0.95, batch_size=256,
policy_kwargs=dict(layers=[256, 256, 256]))
# DDPG Hyperparams:
# NOTE: it works even without action noise
# n_actions = env.action_space.shape[0]
# noise_std = 0.2
# action_noise = NormalActionNoise(mean=np.zeros(n_actions), sigma=noise_std * np.
˓→ones(n_actions))


# model = HER('MlpPolicy', env, DDPG, n_sampled_goal=n_sampled_goal,
# goal_selection_strategy='future',
# verbose=1, buffer_size=int(1e6),
# actor_lr=1e-3, critic_lr=1e-3, action_noise=action_noise,
# gamma=0.95, batch_size=256,
# policy_kwargs=dict(layers=[256, 256, 256]))
model.learn(int(2e5))
model.save('her_sac_highway')
# Load saved model

model = HER.load('her_sac_highway', env=env)
obs = env.reset()
# Evaluate the agent

episode_reward = 0
action, _ = model.predict(obs)
env.render()
episode_reward += reward
if done or info.get('is_success', False):
print("Reward:", episode_reward, "Success?", info.get('is_success', False))
episode_reward = 0.0
obs = env.reset()
1.6.11 Continual Learning
You can also move from learning on one environment to another for continual learning (PPO2 on DemonAttack-v0,
then transferred on SpaceInvaders-v0):
from stable_baselines.common.cmd_util import make_atari_env
# There already exists an environment generator

# that will make and wrap atari environments correctly
env = make_atari_env('DemonAttackNoFrameskip-v4', num_env=8, seed=0)
model = PPO2('CnnPolicy', env, verbose=1)

obs = env.reset()
env.render()
# Close the processes

env.close()
# The number of environments must be identical when changing environments

env = make_atari_env('SpaceInvadersNoFrameskip-v4', num_env=8, seed=0)
1.6. Examples 23

# change env
model.set_env(env)
obs = env.reset()
while True:
env.render()
env.close()
1.6.12 Record a Video
Record a mp4 video (here using a random agent).
Note: It requires ffmpeg or avconv to be installed on the machine.
import gym
from stable_baselines.common.vec_env import VecVideoRecorder, DummyVecEnv
env_id = 'CartPole-v1'
video_folder = 'logs/videos/'
video_length = 100
env = DummyVecEnv([lambda: gym.make(env_id)])
obs = env.reset()
# Record the video starting at the first step

env = VecVideoRecorder(env, video_folder,
record_video_trigger=lambda x: x == 0, video_length=video_
˓→length,
name_prefix="random-agent-{}".format(env_id))
env.reset()
for _ in range(video_length + 1):
action = [env.action_space.sample()]
obs, _, _, _ = env.step(action)
# Save the video
env.close()
1.6.13 Bonus: Make a GIF of a Trained Agent
Note: For Atari games, you need to use a screen recorder such as Kazam. And then convert the video using ffmpeg
import imageio
import numpy as np


model = A2C("MlpPolicy", "LunarLander-v2").learn(100000)
images = []
obs = model.env.reset()
img = model.env.render(mode='rgb_array')
images.append(img)
obs, _, _ ,_ = model.env.step(action)
img = model.env.render(mode='rgb_array')
imageio.mimsave('lander_a2c.gif', [np.array(img) for i, img in enumerate(images) if i

˓→%2 == 0], fps=29)
1.7 Vectorized Environments
Vectorized Environments are a method for stacking multiple independent environments into a single environment.
Instead of training an RL agent on 1 environment per step, it allows us to train it on n environments per step. Because
of this, actions passed to the environment are now a vector (of dimension n). It is the same for observations,
rewards and end of episode signals (dones). In the case of non-array observation spaces such as Dict or Tuple,
where different sub-spaces may have different shapes, the sub-observations are vectors (of dimension n).
Name Box Discrete Dict Tuple Multi Processing

DummyVecEnv X X X X
SubprocVecEnv X X X X X
Note: Vectorized environments are required when using wrappers for frame-stacking or normalization.
Note: When using vectorized environments, the environments are automatically reset at the end of each episode.
Thus, the observation returned for the i-th environment when done[i] is true will in fact be the first observation
of the next episode, not the last observation of the episode that has just terminated. You can access the “real” final
observation of the terminated episode—that is, the one that accompanied the done event provided by the underlying
environment—using the terminal_observation keys in the info dicts returned by the vecenv.
Warning: When using SubprocVecEnv, users must wrap the code in an if __name__ ==
"__main__": if using the forkserver or spawn start method (default on Windows). On Linux, the de-
fault start method is fork which is not thread safe and can create deadlocks.
For more information, see Python’s multiprocessing guidelines.
1.7.1 VecEnv
class stable_baselines.common.vec_env.VecEnv(num_envs, observation_space, ac-

tion_space)
An abstract asynchronous, vectorized environment.
1.7. Vectorized Environments 25

Parameters
• num_envs – (int) the number of environments
• observation_space – (Gym Space) the observation space
• action_space – (Gym Space) the action space
close()
Clean up the environment’s resources.
env_method(method_name, *method_args, indices=None, **method_kwargs)
Call instance methods of vectorized environments.
Parameters
• method_name – (str) The name of the environment method to invoke.
• indices – (list,int) Indices of envs whose method to call
• method_args – (tuple) Any positional arguments to provide in the call
• method_kwargs – (dict) Any keyword arguments to provide in the call
Returns (list) List of items returned by the environment’s method call
get_attr(attr_name, indices=None)
Return attribute from vectorized environment.
Parameters
• attr_name – (str) The name of the attribute whose value to return
• indices – (list,int) Indices of envs to get attribute from
Returns (list) List of values of ‘attr_name’ in all environments
get_images() → Sequence[numpy.ndarray]
Return RGB images from each environment
getattr_depth_check(name, already_found)
Check if an attribute reference is being hidden in a recursive call to __getattr__
Parameters
• name – (str) name of attribute to check for
• already_found – (bool) whether this attribute has already been found in a wrapper
Returns (str or None) name of module whose attribute is being shadowed, if any.
render(mode: str = ’human’)
Gym environment rendering
Parameters mode – the rendering type
reset()
Reset all the environments and return an array of observations, or a tuple of observation arrays.
If step_async is still doing work, that work will be cancelled and step_wait() should not be called until
step_async() is invoked again.
Returns ([int] or [float]) observation
seed(seed: Optional[int] = None) → List[Union[None, int]]
Sets the random seeds for all environments, based on a given seed. Each individual environment will still
get its own seed, by incrementing the given seed.

Parameters seed – (Optional[int]) The random seed. May be None for completely random
seeding.
Returns (List[Union[None, int]]) Returns a list containing the seeds for each individual env.
Note that all list elements may be None, if the env does not return anything when being
seeded.
set_attr(attr_name, value, indices=None)
Set attribute inside vectorized environments.
Parameters
• attr_name – (str) The name of attribute to assign new value
• value – (obj) Value to assign to attr_name
• indices – (list,int) Indices of envs to assign value
Returns (NoneType)
step(actions)
Step the environments with the given action
Parameters actions – ([int] or [float]) the action
Returns ([int] or [float], [float], [bool], dict) observation, reward, done, information
step_async(actions)
Tell all the environments to start taking a step with the given actions. Call step_wait() to get the results of
the step.
You should not call this if a step_async run is already pending.
step_wait()
Wait for the step taken with step_async().
1.7.2 DummyVecEnv
class stable_baselines.common.vec_env.DummyVecEnv(env_fns)
Creates a simple vectorized wrapper for multiple environments, calling each environment in sequence on the
current Python process. This is useful for computationally simple environment such as cartpole-v1, as the
overhead of multiprocess or multithread outweighs the environment computation time. This can also be used
for RL methods that require a vectorized environment, but that you want a single environments to train with.
Parameters env_fns – ([callable]) A list of functions that will create the environments (each
callable returns a Gym.Env instance when called).
close()
Return attribute from vectorized environment (see base class).

render(mode: str = ’human’)

Gym environment rendering. If there are multiple environments then they are tiled together in one im-
age via BaseVecEnv.render(). Otherwise (if self.num_envs == 1), we pass the render call directly to the
underlying environment.
Therefore, some arguments such as mode will have values that are valid only when num_envs == 1.
Parameters mode – The rendering type.
reset()
seed(seed=None)
seeding.
seeded.
Set attribute inside vectorized environments (see base class).
step_async(actions)
the step.
step_wait()
1.7.3 SubprocVecEnv
class stable_baselines.common.vec_env.SubprocVecEnv(env_fns, start_method=None)

Creates a multiprocess vectorized wrapper for multiple environments, distributing each environment to its own
process, allowing significant speed up when the environment is computationally complex.
For performance reasons, if your environment is not IO bound, the number of environments should not exceed
the number of logical cores on your CPU.
Warning: Only ‘forkserver’ and ‘spawn’ start methods are thread-safe, which is important when Tensor-
Flow sessions or other non thread-safe libraries are used in the parent (see issue #217). However, compared
to ‘fork’ they incur a small start-up cost and have restrictions on global variables. With those methods,
users must wrap the code in an if __name__ == "__main__": block. For more information, see the
multiprocessing documentation.
Parameters

• env_fns – ([callable]) A list of functions that will create the environments (each callable
returns a Gym.Env instance when called).
• start_method – (str) method used to start the subprocesses. Must be one of the methods
returned by multiprocessing.get_all_start_methods(). Defaults to ‘forkserver’ on available
platforms, and ‘spawn’ otherwise.
close()
Return attribute from vectorized environment (see base class).
reset()
seed(seed=None)
seeding.
seeded.
Set attribute inside vectorized environments (see base class).
step_async(actions)
the step.
step_wait()
1.7.4 Wrappers
VecFrameStack
class stable_baselines.common.vec_env.VecFrameStack(venv, n_stack)

Frame stacking wrapper for vectorized environment
Parameters

• venv – (VecEnv) the vectorized environment to wrap

• n_stack – (int) Number of frames to stack
close()
reset()
Reset all environments
step_wait()
VecNormalize
class stable_baselines.common.vec_env.VecNormalize(venv, training=True,

norm_obs=True,
norm_reward=True,
clip_obs=10.0, clip_reward=10.0,
gamma=0.99, epsilon=1e-08)
A moving average, normalizing wrapper for vectorized environment.
It is pickleable which will save moving averages and configuration parameters. The wrapped environment venv
is not saved, and must be restored manually with set_venv after being unpickled.
Parameters
• training – (bool) Whether to update or not the moving average
• norm_obs – (bool) Whether to normalize observation or not (default: True)
• norm_reward – (bool) Whether to normalize rewards or not (default: True)
• clip_obs – (float) Max absolute value for observation
• clip_reward – (float) Max value absolute for discounted reward
• gamma – (float) discount factor
• epsilon – (float) To avoid division by zero
get_original_obs() → numpy.ndarray
Returns an unnormalized version of the observations from the most recent step or reset.
get_original_reward() → numpy.ndarray
Returns an unnormalized version of the rewards from the most recent step.
static load(load_path, venv)
Loads a saved VecNormalize object.
Parameters
• load_path – the path to load from.
• venv – the VecEnv to wrap.
Returns (VecNormalize)
load_running_average(path)
Parameters path – (str) path to log dir

Deprecated since version 2.9.0: This function will be removed in a future version
normalize_obs(obs: numpy.ndarray) → numpy.ndarray
Normalize observations using this VecNormalize’s observations statistics. Calling this method does not
update statistics.
normalize_reward(reward: numpy.ndarray) → numpy.ndarray
Normalize rewards using this VecNormalize’s rewards statistics. Calling this method does not update
statistics.
reset()
Reset all environments
save_running_average(path)
Parameters path – (str) path to log dir
Deprecated since version 2.9.0: This function will be removed in a future version
set_venv(venv)
Sets the vector environment to wrap to venv.
Also sets attributes derived from this such as num_env.
Parameters venv – (VecEnv)
step_wait()
Apply sequence of actions to sequence of environments actions -> (observations, rewards, news)
where ‘news’ is a boolean vector indicating whether each element is new.
VecVideoRecorder
class stable_baselines.common.vec_env.VecVideoRecorder(venv, video_folder,

record_video_trigger,
video_length=200,
name_prefix=’rl-video’)
Wraps a VecEnv or VecEnvWrapper object to record rendered image as mp4 video. It requires ffmpeg or avconv
to be installed on the machine.
Parameters
• venv – (VecEnv or VecEnvWrapper)
• video_folder – (str) Where to save videos
• record_video_trigger – (func) Function that defines when to start recording. The
function takes the current number of step, and returns whether we should start recording or
not.
• video_length – (int) Length of recorded videos
• name_prefix – (str) Prefix to the video name
close()
reset()

step_wait()
VecCheckNan
class stable_baselines.common.vec_env.VecCheckNan(venv, raise_exception=False,

warn_once=True, check_inf=True)
NaN and inf checking wrapper for vectorized environment, will raise a warning by default, allowing you to
know from what the NaN of inf originated from.
Parameters
• raise_exception – (bool) Whether or not to raise a ValueError, instead of a UserWarn-
ing
• warn_once – (bool) Whether or not to only warn once.
• check_inf – (bool) Whether or not to check for +inf or -inf as well
reset()
step_async(actions)
the step.
step_wait()
1.8 Using Custom Environments
To use the rl baselines with custom environments, they just need to follow the gym interface. That is to say, your
environment must implement the following methods (and inherits from OpenAI Gym Class):
Note: If you are using images as input, the input values must be in [0, 255] as the observation is normalized (dividing
by 255 to have values in [0, 1]) when using CNN policies.
import gym
from gym import spaces
class CustomEnv(gym.Env):
"""Custom Environment that follows gym interface"""
metadata = {'render.modes': ['human']}


def __init__(self, arg1, arg2, ...):
super(CustomEnv, self).__init__()
# Define action and observation space
# They must be gym.spaces objects
# Example when using discrete actions:
self.action_space = spaces.Discrete(N_DISCRETE_ACTIONS)
# Example for using image as input:
self.observation_space = spaces.Box(low=0, high=255,
shape=(HEIGHT, WIDTH, N_CHANNELS), dtype=np.
˓→uint8)
def step(self, action):

...
return observation, reward, done, info
def reset(self):
...
return observation # reward, done, info can't be included
def render(self, mode='human'):
...
def close (self):
...
Then you can define and train a RL agent with:
# Instantiate the env

# Define and Train the agent
model = A2C('CnnPolicy', env).learn(total_timesteps=1000)
To check that your environment follows the gym interface, please use:
from stable_baselines.common.env_checker import check_env

# It will check your custom environment and output additional warnings if needed
check_env(env)
We have created a colab notebook for a concrete example of creating a custom environment.
You can also find a complete guide online on creating a custom Gym environment.
Optionally, you can also register the environment with gym, that will allow you to create the RL agent in one line (and
use gym.make() to instantiate the env).
In the project, for testing purposes, we use a custom environment named IdentityEnv defined in this file. An
example of how to use it can be found here.
1.9 Custom Policy Network
Stable baselines provides default policy networks (see Policies ) for images (CNNPolicies) and other type of input
features (MlpPolicies).
One way of customising the policy network architecture is to pass arguments when creating the model, using
policy_kwargs parameter:
1.9. Custom Policy Network 33

import gym
import tensorflow as tf
# Custom MLP policy of two layers of size 32 each with tanh activation function
policy_kwargs = dict(act_fun=tf.nn.tanh, net_arch=[32, 32])
# Create the agent
model = PPO2("MlpPolicy", "CartPole-v1", policy_kwargs=policy_kwargs, verbose=1)
# Retrieve the environment
# Train the agent
# Save the agent
model.save("ppo2-cartpole")
del model
# the policy_kwargs are automatically loaded
model = PPO2.load("ppo2-cartpole")
You can also easily define a custom architecture for the policy (or value) network:
Note: Defining a custom policy class is equivalent to passing policy_kwargs. However, it lets you name the
policy and so makes usually the code clearer. policy_kwargs should be rather used when doing hyperparameter
search.
import gym
from stable_baselines.common.policies import FeedForwardPolicy, register_policy


net_arch=[dict(pi=[128, 128, 128],
vf=[128, 128, 128])],

env = DummyVecEnv([lambda: env])
model = A2C(CustomPolicy, env, verbose=1)

# Train the agent
# Save the agent
model.save("a2c-lunar")
del model
# When loading a model with a custom policy
# you MUST pass explicitly the policy when loading the saved model
model = A2C.load("a2c-lunar", policy=CustomPolicy)

Warning: When loading a model with a custom policy, you must pass the custom policy explicitly when loading
the model. (cf previous example)
You can also register your policy, to help with code simplicity: you can refer to your custom policy using a string.
import gym
from stable_baselines.common.policies import FeedForwardPolicy, register_policy


net_arch=[dict(pi=[128, 128, 128],
vf=[128, 128, 128])],
# Register the policy, it will check that the name is not already taken
register_policy('CustomPolicy', CustomPolicy)
# Because the policy is now registered, you can pass

# a string to the agent constructor instead of passing a class
model = A2C(policy='CustomPolicy', env='LunarLander-v2', verbose=1).learn(total_
˓→timesteps=100000)
Deprecated since version 2.3.0: Use net_arch instead of layers parameter to define the network architecture. It
allows to have a greater control.
The net_arch parameter of FeedForwardPolicy allows to specify the amount and size of the hidden layers and
how many of them are shared between the policy network and the value network. It is assumed to be a list with the
following structure:
1. An arbitrary length (zero allowed) number of integers each specifying the number of units in a shared layer. If
the number of ints is zero, there will be no shared layers.
2. An optional dict, to specify the following non-shared layers for the value network and the policy network. It
is formatted like dict(vf=[<value layer sizes>], pi=[<policy layer sizes>]). If it is
missing any of the keys (pi or vf), no non-shared layers (empty list) is assumed.
In short: [<shared layers>, dict(vf=[<non-shared value network layers>],
pi=[<non-shared policy network layers>])].
1.9.1 Examples
Two shared layers of size 128: net_arch=[128, 128]

obs
|
<128>
|
<128>
/ \
action value
Value network deeper than policy network, first layer shared: net_arch=[128, dict(vf=[256, 256])]
1.9. Custom Policy Network 35

obs
|
<128>
/ \
action <256>
|
<256>
|
value
Initially shared then diverging: [128, dict(vf=[256], pi=[16])]

obs
|
<128>
/ \
<16> <256>
| |
action value
The LstmPolicy can be used to construct recurrent policies in a similar way:

class CustomLSTMPolicy(LstmPolicy):
def __init__(self, sess, ob_space, ac_space, n_env, n_steps, n_batch, n_lstm=64,
˓→reuse=False, **_kwargs):
super().__init__(sess, ob_space, ac_space, n_env, n_steps, n_batch, n_lstm,

˓→reuse,
net_arch=[8, 'lstm', dict(vf=[5, 10], pi=[10])],

layer_norm=True, feature_extraction="mlp", **_kwargs)
Here the net_arch parameter takes an additional (mandatory) ‘lstm’ entry within the shared network section. The
LSTM is shared between value network and policy network.
If your task requires even more granular control over the policy architecture, you can redefine the policy directly:
import gym
from stable_baselines.common.policies import ActorCriticPolicy, register_policy,

˓→nature_cnn

# Custom MLP policy of three layers of size 128 each for the actor and 2 layers of 32
˓→for the critic,
# with a nature_cnn feature extractor

class CustomPolicy(ActorCriticPolicy):
def __init__(self, sess, ob_space, ac_space, n_env, n_steps, n_batch, reuse=False,
˓→ **kwargs):
super(CustomPolicy, self).__init__(sess, ob_space, ac_space, n_env, n_steps,

˓→n_batch, reuse=reuse, scale=True)
with tf.variable_scope("model", reuse=reuse):

activ = tf.nn.relu
extracted_features = nature_cnn(self.processed_obs, **kwargs)

extracted_features = tf.layers.flatten(extracted_features)


pi_h = extracted_features
for i, layer_size in enumerate([128, 128, 128]):
pi_h = activ(tf.layers.dense(pi_h, layer_size, name='pi_fc' + str(i)))
pi_latent = pi_h
vf_h = extracted_features
for i, layer_size in enumerate([32, 32]):
vf_h = activ(tf.layers.dense(vf_h, layer_size, name='vf_fc' + str(i)))
value_fn = tf.layers.dense(vf_h, 1, name='vf')
vf_latent = vf_h
self._proba_distribution, self._policy, self.q_value = \

self.pdtype.proba_distribution_from_latent(pi_latent, vf_latent, init_
˓→scale=0.01)
self._value_fn = value_fn
self._setup_init()
def step(self, obs, state=None, mask=None, deterministic=False):

if deterministic:
action, value, neglogp = self.sess.run([self.deterministic_action, self.
˓→value_flat, self.neglogp],
{self.obs_ph: obs})
else:
action, value, neglogp = self.sess.run([self.action, self.value_flat,
˓→self.neglogp],
{self.obs_ph: obs})
return action, value, self.initial_state, neglogp
def proba_step(self, obs, state=None, mask=None):

return self.sess.run(self.policy_proba, {self.obs_ph: obs})
def value(self, obs, state=None, mask=None):

return self.sess.run(self.value_flat, {self.obs_ph: obs})

env = DummyVecEnv([lambda: gym.make('Breakout-v0')])
model = A2C(CustomPolicy, env, verbose=1)

# Train the agent
1.10 Callbacks
A callback is a set of functions that will be called at given stages of the training procedure. You can use callbacks to
access internal state of the RL model during training. It allows one to do monitoring, auto saving, model manipulation,
progress bars, . . .
1.10.1 Custom Callback
To build a custom callback, you need to create a class that derives from BaseCallback. This will give you access
to events (_on_training_start, _on_step) and useful variables (like self.model for the RL model).
1.10. Callbacks 37
You can find two examples of custom callbacks in the documentation: one for saving the best model according to the
training reward (see Examples), and one for logging additional values with Tensorboard (see Tensorboard section).
class CustomCallback(BaseCallback):
"""
A custom callback that derives from ``BaseCallback``.
:param verbose: (int) Verbosity level 0: not output 1: info 2: debug

"""
def __init__(self, verbose=0):
super(CustomCallback, self).__init__(verbose)
# Those variables will be accessible in the callback
# (they are defined in the base class)
# The RL model
# self.model = None # type: BaseRLModel
# An alias for self.model.get_env(), the environment used for training
# self.training_env = None # type: Union[gym.Env, VecEnv, None]
# Number of time the callback was called
# self.n_calls = 0 # type: int
# self.num_timesteps = 0 # type: int
# local and global variables
# self.locals = None # type: Dict[str, Any]
# self.globals = None # type: Dict[str, Any]
# The logger object, used to report things in the terminal
# self.logger = None # type: logger.Logger
# # Sometimes, for event callback, it is useful
# # to have access to the parent object
# self.parent = None # type: Optional[BaseCallback]
def _on_training_start(self) -> None:

"""
This method is called before the first rollout starts.
"""
pass
def _on_rollout_start(self) -> None:

"""
A rollout is the collection of environment interaction
using the current policy.
This event is triggered before collecting new samples.
"""
pass

"""
This method will be called by the model after each call to ènv.step()`.
For child callback (of an ÈventCallback`), this will be called

when the event is triggered.
:return: (bool) If the callback returns False, training is aborted early.

"""
return True
def _on_rollout_end(self) -> None:



"""
This event is triggered before updating the policy.
"""
pass
def _on_training_end(self) -> None:

"""
This event is triggered before exiting the `learn()` method.
"""
pass
Note: self.num_timesteps corresponds to the total number of steps taken in the environment, i.e., it is the number of
environments multiplied by the number of time env.step() was called
You should know that PPO1 and TRPO update self.num_timesteps after each rollout (and not each step) because they
rely on MPI.
For the other algorithms, self.num_timesteps is incremented by n_envs (number of environments) after each call to
env.step()
Note: For off-policy algorithms like SAC, DDPG, TD3 or DQN, the notion of rollout corresponds to the steps
taken in the environment between two updates.
1.10.2 Event Callback
Compared to Keras, Stable Baselines provides a second type of BaseCallback, named EventCallback that is
meant to trigger events. When an event is triggered, then a child callback is called.
As an example, EvalCallback is an EventCallback that will trigger its child callback when there is a new best
model. A child callback is for instance StopTrainingOnRewardThreshold that stops the training if the mean reward
achieved by the RL model is above a threshold.
Note: We recommend to take a look at the source code of EvalCallback and StopTrainingOnRewardThreshold to
have a better overview of what can be achieved with this kind of callbacks.
class EventCallback(BaseCallback):
"""
Base class for triggering callback on event.
:param callback: (Optional[BaseCallback]) Callback that will be called

when an event is triggered.
:param verbose: (int)
"""
def __init__(self, callback: Optional[BaseCallback] = None, verbose: int = 0):
super(EventCallback, self).__init__(verbose=verbose)
self.callback = callback
# Give access to the parent
if callback is not None:
self.callback.parent = self
...
1.10. Callbacks 39
def _on_event(self) -> bool:

if self.callback is not None:
return self.callback()
return True
1.10.3 Callback Collection
Stable Baselines provides you with a set of common callbacks for:

• saving the model periodically (CheckpointCallback)
• evaluating the model periodically and saving the best one (EvalCallback)
• chaining callbacks (CallbackList)
• triggering callback on events (Event Callback, EveryNTimesteps)
• stopping the training early based on a reward threshold (StopTrainingOnRewardThreshold)
CheckpointCallback
Callback for saving a model every save_freq steps, you must specify a log folder (save_path) and optionally a
prefix for the checkpoints (rl_model by default).
from stable_baselines import SAC
from stable_baselines.common.callbacks import CheckpointCallback
# Save a checkpoint every 1000 steps
checkpoint_callback = CheckpointCallback(save_freq=1000, save_path='./logs/',
name_prefix='rl_model')
model = SAC('MlpPolicy', 'Pendulum-v0')

model.learn(2000, callback=checkpoint_callback)
EvalCallback
Evaluate periodically the performance of an agent, using a separate test environment. It will save the best model if
best_model_save_path folder is specified and save the evaluations results in a numpy archive (evaluations.npz)
if log_path folder is specified.
Note: You can pass a child callback via the callback_on_new_best argument. It will be triggered each time
there is a new best model.
import gym

from stable_baselines.common.callbacks import EvalCallback
# Separate evaluation env

eval_env = gym.make('Pendulum-v0')
# Use deterministic actions for evaluation
eval_callback = EvalCallback(eval_env, best_model_save_path='./logs/',


log_path='./logs/', eval_freq=500,
deterministic=True, render=False)

model.learn(5000, callback=eval_callback)
CallbackList
Class for chaining callbacks, they will be called sequentially. Alternatively, you can pass directly a list of callbacks to
the learn() method, it will be converted automatically to a CallbackList.
import gym

from stable_baselines.common.callbacks import CallbackList, CheckpointCallback,
˓→EvalCallback
checkpoint_callback = CheckpointCallback(save_freq=1000, save_path='./logs/')

eval_callback = EvalCallback(eval_env, best_model_save_path='./logs/best_model',
log_path='./logs/results', eval_freq=500)
# Create the callback list
callback = CallbackList([checkpoint_callback, eval_callback])

# Equivalent to:
# model.learn(5000, callback=[checkpoint_callback, eval_callback])
model.learn(5000, callback=callback)
StopTrainingOnRewardThreshold
Stop the training once a threshold in episodic reward (mean episode reward over the evaluations) has been reached
(i.e., when the model is good enough). It must be used with the EvalCallback and use the event triggered by a new
best model.
import gym

from stable_baselines.common.callbacks import EvalCallback,
˓→StopTrainingOnRewardThreshold

# Stop training when the model reaches the reward threshold
callback_on_best = StopTrainingOnRewardThreshold(reward_threshold=-200, verbose=1)
eval_callback = EvalCallback(eval_env, callback_on_new_best=callback_on_best,
˓→verbose=1)
model = SAC('MlpPolicy', 'Pendulum-v0', verbose=1)

# Almost infinite number of timesteps, but the training will stop
# early as soon as the reward threshold is reached
model.learn(int(1e10), callback=eval_callback)
1.10. Callbacks 41
EveryNTimesteps
An Event Callback that will trigger its child callback every n_steps timesteps.
Note: Because of the way PPO1 and TRPO work (they rely on MPI), n_steps is a lower bound between two events.
import gym

from stable_baselines.common.callbacks import CheckpointCallback, EveryNTimesteps
# this is equivalent to defining CheckpointCallback(save_freq=500)

# checkpoint_callback will be triggered every 500 steps
checkpoint_on_event = CheckpointCallback(save_freq=1, save_path='./logs/')
event_callback = EveryNTimesteps(n_steps=500, callback=checkpoint_on_event)
model = PPO2('MlpPolicy', 'Pendulum-v0', verbose=1)
model.learn(int(2e4), callback=event_callback)
Legacy: A functional approach
Warning: This way of doing callbacks is deprecated in favor of the object oriented approach.
A callback function takes the locals() variables and the globals() variables from the model, then returns a
boolean value for whether or not the training should continue.
Thanks to the access to the models variables, in particular _locals["self"], we are able to even change the
parameters of the model without halting the training, or changing the model’s code.
from typing import Dict, Any
def simple_callback(_locals: Dict[str, Any], _globals: Dict[str, Any]) -> bool:

"""
Callback called at each step (for DQN and others) or after n steps (see ACER or
˓→PPO2).
This callback will save the model and stop the training after the first call.
:param _locals: (Dict[str, Any])

:param _globals: (Dict[str, Any])
:return: (bool) If your callback returns False, training is aborted early.
"""
print("callback called")
# Save the model
_locals["self"].save("saved_model")
# If you want to continue training, the callback must return True.
# return True # returns True, training continues.
print("stop training")
return False # returns False, training stops.

model = PPO2('MlpPolicy', 'CartPole-v1')

model.learn(2000, callback=simple_callback)
class stable_baselines.common.callbacks.BaseCallback(verbose: int = 0)

Base class for callback.
Parameters verbose – (int)
init_callback(model: BaseRLModel) → None
Initialize the callback by saving references to the RL model and the training environment for convenience.
on_step() → bool
This method will be called by the model after each call to env.step().
For child callback (of an EventCallback), this will be called when the event is triggered.
Returns (bool) If the callback returns False, training is aborted early.
update_locals(locals_: Dict[str, Any]) → None
Updates the local variables of the training process
For reference to which variables are accessible, check each individual algorithm’s documentation :param
locals_: (Dict[str, Any]) current local variables
class stable_baselines.common.callbacks.CallbackList(callbacks:
List[stable_baselines.common.callbacks.BaseCallback])
Class for chaining callbacks.
Parameters callbacks – (List[BaseCallback]) A list of callbacks that will be called sequentially.
class stable_baselines.common.callbacks.CheckpointCallback(save_freq: int,
save_path: str,
name_prefix=’rl_model’,
verbose=0)
Callback for saving a model every save_freq steps
Parameters
• save_freq – (int)
• save_path – (str) Path to the folder where the model will be saved.
• name_prefix – (str) Common prefix to the saved models
class stable_baselines.common.callbacks.ConvertCallback(callback, verbose=0)
Convert functional callback (old-style) to object.
Parameters
• callback – (Callable)
• verbose – (int)
1.10. Callbacks 43
class stable_baselines.common.callbacks.EvalCallback(eval_env:
Union[gym.core.Env, sta-
ble_baselines.common.vec_env.base_vec_env.VecEnv],
callback_on_new_best: Op-
tional[stable_baselines.common.callbacks.BaseCallback
= None, n_eval_episodes:
int = 5, eval_freq: int =
10000, log_path: str = None,
best_model_save_path: str =
None, deterministic: bool =
True, render: bool = False,
verbose: int = 1)
Callback for evaluating an agent.
Parameters
• eval_env – (Union[gym.Env, VecEnv]) The environment used for initialization
• callback_on_new_best – (Optional[BaseCallback]) Callback to trigger when there is
a new best model according to the mean_reward
• n_eval_episodes – (int) The number of episodes to test the agent
• eval_freq – (int) Evaluate the agent every eval_freq call of the callback.
• log_path – (str) Path to a folder where the evaluations (evaluations.npz) will be saved. It
will be updated at each evaluation.
• best_model_save_path – (str) Path to a folder where the best model according to
performance on the eval env will be saved.
• deterministic – (bool) Whether the evaluation should use a stochastic or deterministic
actions.
• render – (bool) Whether to render or not the environment during evaluation
class stable_baselines.common.callbacks.EventCallback(callback: Op-
tional[stable_baselines.common.callbacks.BaseCallba
= None, verbose: int = 0)
Base class for triggering callback on event.
Parameters
• callback – (Optional[BaseCallback]) Callback that will be called when an event is trig-
gered.
init_callback(model: BaseRLModel) → None
Initialize the callback by saving references to the RL model and the training environment for convenience.
class stable_baselines.common.callbacks.EveryNTimesteps(n_steps: int, callback: sta-
ble_baselines.common.callbacks.BaseCallback)
Trigger a callback every n_steps timesteps
Parameters
• n_steps – (int) Number of timesteps between two trigger.
• callback – (BaseCallback) Callback that will be called when the event is triggered.

class stable_baselines.common.callbacks.StopTrainingOnRewardThreshold(reward_threshold:
float,
ver-
bose:
int =
0)
Stop the training once a threshold in episodic reward has been reached (i.e. when the model is good enough).
It must be used with the EvalCallback.
Parameters
• reward_threshold – (float) Minimum expected reward per episode to stop training.
1.11 Tensorboard Integration
1.11.1 Basic Usage
To use Tensorboard with the rl baselines, you simply need to define a log location for the RL agent:
import gym
model = A2C('MlpPolicy', 'CartPole-v1', verbose=1, tensorboard_log="./a2c_cartpole_

˓→tensorboard/")
Or after loading an existing model (by default the log path is not saved):
import gym

env = DummyVecEnv([lambda: env]) # The algorithms require a vectorized environment
˓→to run
model = A2C.load("./a2c_cartpole.pkl", env=env, tensorboard_log="./a2c_cartpole_

˓→tensorboard/")
You can also define custom logging name when training (by default it is the algorithm name)
import gym
model = A2C('MlpPolicy', 'CartPole-v1', verbose=1, tensorboard_log="./a2c_cartpole_

˓→tensorboard/")
model.learn(total_timesteps=10000, tb_log_name="first_run")
# Pass reset_num_timesteps=False to continue the training curve in tensorboard
# By default, it will create a new curve
1.11. Tensorboard Integration 45


model.learn(total_timesteps=10000, tb_log_name="second_run", reset_num_
˓→timesteps=False)
model.learn(total_timesteps=10000, tb_log_name="thrid_run", reset_num_timesteps=False)
Once the learn function is called, you can monitor the RL agent during or after the training, with the following bash
command:
tensorboard --logdir ./a2c_cartpole_tensorboard/
you can also add past logging folders:
tensorboard --logdir ./a2c_cartpole_tensorboard/;./ppo2_cartpole_tensorboard/
It will display information such as the model graph, the episode reward, the model losses, the observation and other
parameter unique to some models.

1.11.2 Logging More Values
Using a callback, you can easily log more values with TensorBoard. Here is a simple example on how to log both
additional tensor or arbitrary scalar value:
import numpy as np

model = SAC("MlpPolicy", "Pendulum-v0", tensorboard_log="/tmp/sac/", verbose=1)
class TensorboardCallback(BaseCallback):
"""
Custom callback for plotting additional values in tensorboard.
"""
def __init__(self, verbose=0):
self.is_tb_set = False
super(TensorboardCallback, self).__init__(verbose)

# Log additional tensor
if not self.is_tb_set:
with self.model.graph.as_default():
tf.summary.scalar('value_target', tf.reduce_mean(self.model.value_
˓→target))
self.model.summary = tf.summary.merge_all()
self.is_tb_set = True
# Log scalar value (here a random variable)
value = np.random.random()
summary = tf.Summary(value=[tf.Summary.Value(tag='random_value', simple_
˓→value=value)])
self.locals['writer'].add_summary(summary, self.num_timesteps)
return True
model.learn(50000, callback=TensorboardCallback())
1.11. Tensorboard Integration 47

1.11.3 Legacy Integration
All the information displayed in the terminal (default logging) can be also logged in tensorboard. For that, you need
to define several environment variables:
# formats are comma-separated, but for tensorboard you only need the last one
# stdout -> terminal
export OPENAI_LOG_FORMAT='stdout,log,csv,tensorboard'
export OPENAI_LOGDIR=path/to/tensorboard/data
and to configure the logger using:
from stable_baselines.logger import configure
configure()
Then start tensorboard with:
tensorboard --logdir=$OPENAI_LOGDIR
1.12 RL Baselines Zoo
RL Baselines Zoo. is a collection of pre-trained Reinforcement Learning agents using Stable-Baselines. It also
provides basic scripts for training, evaluating agents, tuning hyperparameters and recording videos.
Goals of this repository:
1. Provide a simple interface to train and enjoy RL agents
2. Benchmark the different Reinforcement Learning algorithms
3. Provide tuned hyperparameters for each environment and RL algorithm
4. Have fun with the trained agents!
1.12.1 Installation
1. Install dependencies
apt-get install swig cmake libopenmpi-dev zlib1g-dev ffmpeg

pip install stable-baselines box2d box2d-kengz pyyaml pybullet optuna pytablewriter
2. Clone the repository:
git clone https://github.com/araffin/rl-baselines-zoo
1.12.2 Train an Agent
The hyperparameters for each environment are defined in hyperparameters/algo_name.yml.

If the environment exists in this file, then you can train an agent using:
python train.py --algo algo_name --env env_id
For example (with tensorboard support):

python train.py --algo ppo2 --env CartPole-v1 --tensorboard-log /tmp/stable-baselines/
Train for multiple environments (with one call) and with tensorboard logging:
python train.py --algo a2c --env MountainCar-v0 CartPole-v1 --tensorboard-log /tmp/

˓→stable-baselines/
Continue training (here, load pretrained agent for Breakout and continue training for 5000 steps):
python train.py --algo a2c --env BreakoutNoFrameskip-v4 -i trained_agents/a2c/

˓→BreakoutNoFrameskip-v4.pkl -n 5000
1.12.3 Enjoy a Trained Agent
If the trained agent exists, then you can see it in action using:
python enjoy.py --algo algo_name --env env_id
For example, enjoy A2C on Breakout during 5000 timesteps:
python enjoy.py --algo a2c --env BreakoutNoFrameskip-v4 --folder trained_agents/ -n

˓→5000
1.12.4 Hyperparameter Optimization
We use Optuna for optimizing the hyperparameters.

Tune the hyperparameters for PPO2, using a random sampler and median pruner, 2 parallels jobs, with a budget of
1000 trials and a maximum of 50000 steps:
python train.py --algo ppo2 --env MountainCar-v0 -n 50000 -optimize --n-trials 1000 --
˓→n-jobs 2 \
--sampler random --pruner median
1.12.5 Colab Notebook: Try it Online!
You can train agents online using Google colab notebook.
Note: You can find more information about the rl baselines zoo in the repo README. For instance, how to record a
video of a trained agent.
1.13 Pre-Training (Behavior Cloning)
With the .pretrain() method, you can pre-train RL policies using trajectories from an expert, and therefore
accelerate training.
Behavior Cloning (BC) treats the problem of imitation learning, i.e., using expert demonstrations, as a supervised
learning problem. That is to say, given expert trajectories (observations-actions pairs), the policy network is trained
1.13. Pre-Training (Behavior Cloning) 49

to reproduce the expert behavior: for a given observation, the action taken by the policy must be the one taken by the
expert.
Expert trajectories can be human demonstrations, trajectories from another controller (e.g. a PID controller) or trajec-
tories from a trained RL agent.
Note: Only Box and Discrete spaces are supported for now for pre-training a model.
Note: Images datasets are treated a bit differently as other datasets to avoid memory issues. The images from the
expert demonstrations must be located in a folder, not in the expert numpy archive.
1.13.1 Generate Expert Trajectories
Here, we are going to train a RL model and then generate expert trajectories using this agent.
Note that in practice, generating expert trajectories usually does not require training an RL agent.
The following example is only meant to demonstrate the pretrain() feature.
However, we recommend users to take a look at the code of the generate_expert_traj() function (located in
gail/dataset/ folder) to learn about the data structure of the expert dataset (see below for an overview) and how
to record trajectories.

from stable_baselines.gail import generate_expert_traj
model = DQN('MlpPolicy', 'CartPole-v1', verbose=1)

# Train a DQN agent for 1e5 timesteps and generate 10 trajectories
# data will be saved in a numpy archive named èxpert_cartpole.npz`
generate_expert_traj(model, 'expert_cartpole', n_timesteps=int(1e5), n_episodes=10)
Here is an additional example when the expert controller is a callable, that is passed to the function instead of a RL
model. The idea is that this callable can be a PID controller, asking a human player, . . .
import gym
env = gym.make("CartPole-v1")
# Here the expert is a random agent
# but it can be any python function, e.g. a PID controller
def dummy_expert(_obs):
"""
Random agent. It samples actions randomly
from the action space of the environment.
:param _obs: (np.ndarray) Current observation

:return: (np.ndarray) action taken by the expert
"""
return env.action_space.sample()
# Data will be saved in a numpy archive named èxpert_cartpole.npz`
# when using something different than an RL expert,
# you must pass the environment object explicitly
generate_expert_traj(dummy_expert, 'dummy_expert_cartpole', env, n_episodes=10)

1.13.2 Pre-Train a Model using Behavior Cloning
Using the expert_cartpole.npz dataset generated with the previous script.

from stable_baselines.gail import ExpertDataset
# Using only one expert trajectory
# you can specify `traj_limitation=-1` for using the whole dataset
dataset = ExpertDataset(expert_path='expert_cartpole.npz',
traj_limitation=1, batch_size=128)
model = PPO2('MlpPolicy', 'CartPole-v1', verbose=1)

# Pretrain the PPO2 model
model.pretrain(dataset, n_epochs=1000)
# As an option, you can train the RL agent

# model.learn(int(1e5))
# Test the pre-trained model

obs = env.reset()
reward_sum = 0.0
obs, reward, done, _ = env.step(action)
reward_sum += reward
env.render()
if done:
print(reward_sum)
reward_sum = 0.0
obs = env.reset()
env.close()
1.13.3 Data Structure of the Expert Dataset
The expert dataset is a .npz archive. The data is saved in python dictionary format with keys: actions,
episode_returns, rewards, obs, episode_starts.
In case of images, obs contains the relative path to the images.
obs, actions: shape (N * L, ) + S
where N = # episodes, L = episode length and S is the environment observation/action space.
S = (1, ) for discrete space
class stable_baselines.gail.ExpertDataset(expert_path=None, traj_data=None,
train_fraction=0.7, batch_size=64,
traj_limitation=-1, randomize=True, ver-
bose=1, sequential_preprocessing=False)
Dataset for using behavior cloning or GAIL.
The structure of the expert dataset is a dict, saved as an “.npz” archive. The dictionary contains the keys
‘actions’, ‘episode_returns’, ‘rewards’, ‘obs’ and ‘episode_starts’. The corresponding values have data concate-
nated across episode: the first axis is the timestep, the remaining axes index into the data. In case of images,
‘obs’ contains the relative path to the images, to enable space saving from image compression.
1.13. Pre-Training (Behavior Cloning) 51

Parameters
• expert_path – (str) The path to trajectory data (.npz file). Mutually exclusive with
traj_data.
• traj_data – (dict) Trajectory data, in format described above. Mutually exclusive with
expert_path.
• train_fraction – (float) the train validation split (0 to 1) for pre-training using behav-
ior cloning (BC)
• batch_size – (int) the minibatch size for behavior cloning
• traj_limitation – (int) the number of trajectory to use (if -1, load all)
• randomize – (bool) if the dataset should be shuffled
• verbose – (int) Verbosity
• sequential_preprocessing – (bool) Do not use subprocess to preprocess the data
(slower but use less memory for the CI)
get_next_batch(split=None)
Get the batch from the dataset.
Parameters split – (str) the type of data split (can be None, ‘train’, ‘val’)
Returns (np.ndarray, np.ndarray) inputs and labels
init_dataloader(batch_size)
Initialize the dataloader used by GAIL.
Parameters batch_size – (int)
log_info()
Log the information of the dataset.
plot()
Show histogram plotting of the episode returns
class stable_baselines.gail.DataLoader(indices, observations, actions, batch_size,
n_workers=1, infinite_loop=True,
max_queue_len=1, shuffle=False,
start_process=True, backend=’threading’, se-
quential=False, partial_minibatch=True)
A custom dataloader to preprocessing observations (including images) and feed them to the network.
Original code for the dataloader from https://github.com/araffin/robotics-rl-srl (MIT licence) Authors: Antonin
Raffin, René Traoré, Ashley Hill
Parameters
• indices – ([int]) list of observations indices
• observations – (np.ndarray) observations or images path
• actions – (np.ndarray) actions
• batch_size – (int) Number of samples per minibatch
• n_workers – (int) number of preprocessing worker (for loading the images)
• infinite_loop – (bool) whether to have an iterator that can be reset
• max_queue_len – (int) Max number of minibatches that can be preprocessed at the same
time

• shuffle – (bool) Shuffle the minibatch after each epoch

• start_process – (bool) Start the preprocessing process (default: True)
• backend – (str) joblib backend (one of ‘multiprocessing’, ‘sequential’, ‘threading’ or
‘loky’ in newest versions)
• sequential – (bool) Do not use subprocess to preprocess the data (slower but use less
memory for the CI)
• partial_minibatch – (bool) Allow partial minibatches (minibatches with a number of
element lesser than the batch_size)
sequential_next()
Sequential version of the pre-processing.
start_process()
Start preprocessing process
stable_baselines.gail.generate_expert_traj(model, save_path=None, env=None,
n_timesteps=0, n_episodes=100, im-
age_folder=’recorded_images’)
Train expert controller (if needed) and record expert trajectories.
Note: only Box and Discrete spaces are supported for now.
Parameters
• model – (RL model or callable) The expert model, if it needs to be trained, then you need
to pass n_timesteps > 0.
• save_path – (str) Path without the extension where the expert dataset will be saved (ex:
‘expert_cartpole’ -> creates ‘expert_cartpole.npz’). If not specified, it will not save, and just
return the generated expert trajectories. This parameter must be specified for image-based
environments.
• env – (gym.Env) The environment, if not defined then it tries to use the model environment.
• n_timesteps – (int) Number of training timesteps
• n_episodes – (int) Number of trajectories (episodes) to record
• image_folder – (str) When using images, folder that will be used to record images.
Returns (dict) the generated expert trajectories.
1.14 Dealing with NaNs and infs
During the training of a model on a given environment, it is possible that the RL model becomes completely corrupted
when a NaN or an inf is given or returned from the RL model.
1.14.1 How and why?
The issue arises then NaNs or infs do not crash, but simply get propagated through the training, until all the floating
point number converge to NaN or inf. This is in line with the IEEE Standard for Floating-Point Arithmetic (IEEE 754)
standard, as it says:
1.14. Dealing with NaNs and infs 53

Note:
Five possible exceptions can occur:
√
• Invalid operation ( −1, inf ×1, NaN mod 1, . . . ) return NaN
• Division by zero:
– if the operand is not zero (1/0, −2/0, . . . ) returns ± inf
– if the operand is zero (0/0) returns signaling NaN
• Overflow (exponent too high to represent) returns ± inf
• Underflow (exponent too low to represent) returns 0
• Inexact (not representable exactly in base 2, eg: 1/5) returns the rounded value (ex: assert (1/5) *
3 == 0.6000000000000001)
And of these, only Division by zero will signal an exception, the rest will propagate invalid values quietly.
In python, dividing by zero will indeed raise the exception: ZeroDivisionError: float division by
zero, but ignores the rest.
The default in numpy, will warn: RuntimeWarning: invalid value encountered but will not halt the
code.
And the worst of all, Tensorflow will not signal anything
import numpy as np
print("tensorflow test:")
a = tf.constant(1.0)
b = tf.constant(0.0)
c = a / b
sess = tf.Session()
val = sess.run(c) # this will be quiet
print(val)
sess.close()
print("\r\nnumpy test:")
a = np.float64(1.0)
b = np.float64(0.0)
val = a / b # this will warn
print(val)
print("\r\npure python test:")
a = 1.0
b = 0.0
val = a / b # this will raise an exception and halt.
print(val)
Unfortunately, most of the floating point operations are handled by Tensorflow and numpy, meaning you might get
little to no warning when a invalid value occurs.

1.14.2 Numpy parameters
Numpy has a convenient way of dealing with invalid value: numpy.seterr, which defines for the python process, how
it should handle floating point error.
import numpy as np
np.seterr(all='raise') # define before your code.
print("numpy test:")
a = np.float64(1.0)
b = np.float64(0.0)
val = a / b # this will now raise an exception instead of a warning.
print(val)
but this will also avoid overflow issues on floating point numbers:
import numpy as np
print("numpy overflow test:")
a = np.float64(10)
b = np.float64(1000)
val = a ** b # this will now raise an exception
print(val)
but will not avoid the propagation issues:

import numpy as np
print("numpy propagation test:")
a = np.float64('NaN')
b = np.float64(1.0)
val = a + b # this will neither warn nor raise anything
print(val)
1.14.3 Tensorflow parameters
Tensorflow can add checks for detecting and dealing with invalid value: tf.add_check_numerics_ops and
tf.check_numerics, however they will add operations to the Tensorflow graph and raise the computation time.
print("tensorflow test:")
a = tf.constant(1.0)
c = a / b
check_nan = tf.add_check_numerics_ops() # add after your graph definition.


sess = tf.Session()
val, _ = sess.run([c, check_nan]) # this will now raise an exception
print(val)
sess.close()
but this will also avoid overflow issues on floating point numbers:
print("tensorflow overflow test:")
check_nan = [] # the list of check_numerics operations
a = tf.constant(10)
b = tf.constant(1000)
c = a ** b
check_nan.append(tf.check_numerics(c, "")) # check the 'c' operations
sess = tf.Session()
val, _ = sess.run([c] + check_nan) # this will now raise an exception
print(val)
sess.close()
and catch propagation issues:
print("tensorflow propagation test:")
check_nan = [] # the list of check_numerics operations
a = tf.constant('NaN')
c = a + b
check_nan.append(tf.check_numerics(c, "")) # check the 'c' operations
sess = tf.Session()
val, _ = sess.run([c] + check_nan) # this will now raise an exception
print(val)
sess.close()
1.14.4 VecCheckNan Wrapper
In order to find when and from where the invalid value originated from, stable-baselines comes with a VecCheckNan
wrapper.
It will monitor the actions, observations, and rewards, indicating what action or observation caused it and from what.
import gym
from gym import spaces
import numpy as np


from stable_baselines.common.vec_env import DummyVecEnv, VecCheckNan
class NanAndInfEnv(gym.Env):
"""Custom Environment that raised NaNs and Infs"""
metadata = {'render.modes': ['human']}
def __init__(self):
super(NanAndInfEnv, self).__init__()
self.action_space = spaces.Box(low=-np.inf, high=np.inf, shape=(1,), dtype=np.
˓→float64)
self.observation_space = spaces.Box(low=-np.inf, high=np.inf, shape=(1,),

˓→dtype=np.float64)
def step(self, _action):

randf = np.random.rand()
if randf > 0.99:
obs = float('NaN')
elif randf > 0.98:
obs = float('inf')
else:
obs = randf
return [obs], 0.0, False, {}
def reset(self):
return [0.0]
def render(self, mode='human', close=False):

pass
# Create environment
env = DummyVecEnv([lambda: NanAndInfEnv()])
env = VecCheckNan(env, raise_exception=True)
# Instantiate the agent

model = PPO2('MlpPolicy', env)
# Train the agent

model.learn(total_timesteps=int(2e5)) # this will crash explaining that the invalid
˓→value originated from the environment.
1.14.5 RL Model hyperparameters
Depending on your hyperparameters, NaN can occurs much more often. A great example of this: https://github.com/
hill-a/stable-baselines/issues/340
Be aware, the hyperparameters given by default seem to work in most cases, however your environment might not
play nice with them. If this is the case, try to read up on the effect each hyperparameters has on the model, so that you
can try and tune them to get a stable model. Alternatively, you can try automatic hyperparameter tuning (included in
the rl zoo).

1.14.6 Missing values from datasets
If your environment is generated from an external dataset, do not forget to make sure your dataset does not contain
NaNs. As some datasets will sometimes fill missing values with NaNs as a surrogate value.
Here is some reading material about finding NaNs: https://pandas.pydata.org/pandas-docs/stable/user_guide/missing_
data.html
And filling the missing values with something else (imputation): https://towardsdatascience.com/
how-to-handle-missing-data-8646b18db0d4
1.15 On saving and loading
Stable baselines stores both neural network parameters and algorithm-related parameters such as exploration schedule,
number of environments and observation/action space. This allows continual learning and easy use of trained agents
without training, but it is not without its issues. Following describes two formats used to save agents in stable baselines,
their pros and shortcomings.
Terminology used in this page:
• parameters refer to neural network parameters (also called “weights”). This is a dictionary mapping Tensorflow
variable name to a NumPy array.
• data refers to RL algorithm parameters, e.g. learning rate, exploration schedule, action/observation space. These
depend on the algorithm used. This is a dictionary mapping classes variable names their values.
1.15.1 Cloudpickle (stable-baselines<=2.7.0)
Original stable baselines save format. Data and parameters are bundled up into a tuple (data, parameters) and
then serialized with cloudpickle library (essentially the same as pickle).
This save format is still available via an argument in model save function in stable-baselines versions above v2.7.0 for
backwards compatibility reasons, but its usage is discouraged.
Pros:
• Easy to implement and use.
• Works with almost any type of Python object, including functions.
Cons:
• Pickle/Cloudpickle is not designed for long-term storage or sharing between Python version.
• If one object in file is not readable (e.g. wrong library version), then reading the rest of the file is difficult.
• Python-specific format, hard to read stored files from other languages.
If part of a saved model becomes unreadable for any reason (e.g. different Tensorflow versions), then it may be tricky
to restore any of the model. For this reason another save format was designed.
1.15.2 Zip-archive (stable-baselines>2.7.0)
A zip-archived JSON dump and NumPy zip archive of the arrays. The data dictionary (class parameters) is stored as a
JSON file, model parameters are serialized with numpy.savez function and these two files are stored under a single
.zip archive.

Any objects that are not JSON serializable are serialized with cloudpickle and stored as base64-encoded string in the
JSON file, along with some information that was stored in the serialization. This allows inspecting stored objects
without deserializing the object itself.
This format allows skipping elements in the file, i.e. we can skip deserializing objects that are broken/non-serializable.
This can be done via custom_objects argument to load functions.
This is the default save format in stable baselines versions after v2.7.0.
File structure:
saved_model.zip/
data JSON file of class-parameters (dictionary)
parameter_list JSON file of model parameters and their ordering (list)
parameters Bytes from numpy.savez (a zip file of the numpy arrays). ...
... Being a zip-archive itself, this object can also be opened ...
... as a zip-archive and browsed.
Pros:
• More robust to unserializable objects (one bad object does not break everything).
• Saved file can be inspected/extracted with zip-archive explorers and by other languages.
Cons:
• More complex implementation.
• Still relies partly on cloudpickle for complex objects (e.g. custom functions).
1.16 Exporting models
After training an agent, you may want to deploy/use it in an other language or framework, like PyTorch or tensorflowjs.
Stable Baselines does not include tools to export models to other frameworks, but this document aims to cover parts
that are required for exporting along with more detailed stories from users of Stable Baselines.
1.16.1 Background
In Stable Baselines, the controller is stored inside policies which convert observations into actions. Each learning
algorithm (e.g. DQN, A2C, SAC) contains one or more policies, some of which are only used for training. An easy
way to find the policy is to check the code for the predict function of the agent: This function should only call one
policy with simple arguments.
Policies hold the necessary Tensorflow placeholders and tensors to do the inference (i.e. predict actions), so it is
enough to export these policies to do inference in an another framework.
Note: Learning algorithms also may contain other Tensorflow placeholders, that are used for training only and are
not required for inference.
Warning: When using CNN policies, the observation is normalized internally (dividing by 255 to have values in
[0, 1])
1.16. Exporting models 59

1.16.2 Export to PyTorch
A known working solution is to use get_parameters function to obtain model parameters, construct the network
manually in PyTorch and assign parameters correctly.
Warning: PyTorch and Tensorflow have internal differences with e.g. 2D convolutions (see discussion linked
below).
See discussion #372 for details.
1.16.3 Export to C++
Tensorflow, which is the backbone of Stable Baselines, is fundamentally a C/C++ library despite being most commonly
accessed through the Python frontend layer. This design choice means that the models created at Python level should
generally be fully compliant with the respective C++ version of Tensorflow.
Warning: It is advisable not to mix-and-match different versions of Tensorflow libraries, particularly in terms
of the state. Moving computational graphs is generally more forgiving. As a matter of fact, mentioned below
PPO_CPP project uses graphs generated with Python Tensorflow 1.x in C++ Tensorflow 2 version.
Stable Baselines comes very handily when hoping to migrate a computational graph and/or a state (weights) as the
existing algorithms define most of the necessary computations for you so you don’t need to recreate the core of the
algorithms again. This is exactly the idea that has been used in the PPO_CPP project, which executes the training
at the C++ level for the sake of computational efficiency. The graphs are exported from Stable Baselines’ PPO2
implementation through tf.train.export_meta_graph function. Alternatively, and perhaps more commonly,
you could use the C++ layer only for inference. That could be useful as a deployment step of server backends or
optimization for more limited devices.
Warning: As a word of caution, C++-level APIs are more imperative than their Python counterparts or more
plainly speaking: cruder. This is particularly apparent in Tensorflow 2.0 where the declarativeness of Autograph
exists only at Python level. The C++ counterpart still operates on Session objects’ use, which are known from
earlier versions of Tensorflow. In our use case, availability of graphs utilized by Session depends on the use of
tf.function decorators. However, as of November 2019, Stable Baselines still uses Tensorflow 1.x in the main
version which is slightly easier to use in the context of the C++ portability.
1.16.4 Export to tensorflowjs / tfjs
Can be done via Tensorflow’s simple_save function and tensorflowjs_converter.

See discussion #474 for details.
1.16.5 Export to Java
Can be done via Tensorflow’s simple_save function.

See this discussion for details.

1.16.6 Manual export
You can also manually export required parameters (weights) and construct the network in your desired framework, as
done with the PyTorch example above.
You can access parameters of the model via agents’ get_parameters function. If you use default policies, you
can find the architecture of the networks in source for policies. Otherwise, for DQN/SAC/DDPG or TD3 you need to
check the policies.py file located in their respective folders.
1.17 Base RL Class
Common interface for all the RL algorithms

class stable_baselines.common.base_class.BaseRLModel(policy, env, verbose=0,
*, requires_vec_env,
policy_base, pol-
icy_kwargs=None, seed=None,
n_cpu_tf_sess=None)
The base RL model
Parameters
• policy – (BasePolicy) Policy object
• env – (Gym environment) The environment to learn from (if registered in Gym, can be str.
Can be None for loading trained models)
• verbose – (int) the verbosity level: 0 none, 1 training information, 2 tensorflow debug
• requires_vec_env – (bool) Does this model require a vectorized environment
• policy_base – (BasePolicy) the base policy used by this method
• policy_kwargs – (dict) additional arguments to be passed to the policy on creation
• seed – (int) Seed for the pseudo-random generators (python, numpy, tensorflow). If None
(default), use random seed. Note that if you want completely deterministic results, you must
set n_cpu_tf_sess to 1.
• n_cpu_tf_sess – (int) The number of threads for TensorFlow operations If None, the
number of cpu of the current machine will be used.
action_probability(observation, state=None, mask=None, actions=None, logp=False)
If actions is None, then get the model’s action probability distribution from a given observation.
Depending on the action space the output is:
• Discrete: probability for each possible action
• Box: mean and standard deviation of the action output
However if actions is not None, this function will return the probability that the given actions are taken
with the given parameters (observation, state, . . . ) on this model. For discrete action spaces, it returns
the probability mass; for continuous action spaces, the probability density. This is since the probability
mass will always be zero in continuous spaces, see http://blog.christianperone.com/2019/01/ for a good
explanation
Parameters
• observation – (np.ndarray) the input observation
• state – (np.ndarray) The last states (can be None, used in recurrent policies)
1.17. Base RL Class 61

• mask – (np.ndarray) The last masks (can be None, used in recurrent policies)
• actions – (np.ndarray) (OPTIONAL) For calculating the likelihood that the given ac-
tions are chosen by the model for each of the given parameters. Must have the same
number of actions and observations. (set to None to return the complete action probability
distribution)
• logp – (bool) (OPTIONAL) When specified with actions, returns probability in log-
space. This has no effect if actions is None.
Returns (np.ndarray) the model’s (log) action probability
get_env()
returns the current environment (can be None if not defined)
Returns (Gym Environment) The current environment
get_parameter_list()
Get tensorflow Variables of model’s parameters
This includes all variables necessary for continuing training (saving / loading).
Returns (list) List of tensorflow Variables
get_parameters()
Get current model parameters as dictionary of variable name -> ndarray.
Returns (OrderedDict) Dictionary of variable name -> ndarray of model’s parameters.
get_vec_normalize_env() → Optional[stable_baselines.common.vec_env.vec_normalize.VecNormalize]
Return the VecNormalize wrapper of the training env if it exists.
Returns Optional[VecNormalize] The VecNormalize env.
learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’run’, re-
set_num_timesteps=True)
Return a trained model.
Parameters
• total_timesteps – (int) The total number of samples to train on
• callback – (Union[callable, [callable], BaseCallback]) function called at every steps
with state of the algorithm. It takes the local and global variables. If it returns False,
training is aborted. When the callback inherits from BaseCallback, you will have access
to additional stages of the training (training start/end), please read the documentation for
more details.
• log_interval – (int) The number of timesteps before logging.
• tb_log_name – (str) the name of the run for tensorboard log
• reset_num_timesteps – (bool) whether or not to reset the current timestep number
(used in logging)
Returns (BaseRLModel) the trained model
classmethod load(load_path, env=None, custom_objects=None, **kwargs)
Load the model from file
Parameters
• load_path – (str or file-like) the saved parameter location
• env – (Gym Environment) the new environment to run the loaded model on (can be None
if you only need prediction from a trained model)

• custom_objects – (dict) Dictionary of objects to replace upon loading. If a variable

is present in this dictionary as a key, it will not be deserialized and the corresponding item
will be used instead. Similar to custom_objects in keras.models.load_model. Useful when
you have an object in file that can not be deserialized.
• kwargs – extra arguments to change the model when loading
load_parameters(load_path_or_dict, exact_match=True)
Load model parameters from a file or a dictionary
Dictionary keys should be tensorflow variable names, which can be obtained with get_parameters
function. If exact_match is True, dictionary should contain keys for all model’s parameters, otherwise
RunTimeError is raised. If False, only variables included in the dictionary will be updated.
This does not load agent’s hyper-parameters.
Warning: This function does not update trainer/optimizer variables (e.g. momentum). As such
training after using this function may lead to less-than-optimal results.
Parameters
• load_path_or_dict – (str or file-like or dict) Save parameter location or dict of pa-
rameters as variable.name -> ndarrays to be loaded.
• exact_match – (bool) If True, expects load dictionary to contain keys for all variables
in the model. If False, loads parameters only for variables mentioned in the dictionary.
Defaults to True.
predict(observation, state=None, mask=None, deterministic=False)

Get the model’s action from an observation
Parameters
• deterministic – (bool) Whether or not to return deterministic actions.
Returns (np.ndarray, np.ndarray) the model’s action and the next state (used in recurrent poli-
cies)
pretrain(dataset, n_epochs=10, learning_rate=0.0001, adam_epsilon=1e-08, val_interval=None)
Pretrain a model using behavior cloning: supervised learning given an expert dataset.
NOTE: only Box and Discrete spaces are supported for now.
Parameters
• dataset – (ExpertDataset) Dataset manager
• n_epochs – (int) Number of iterations on the training set
• learning_rate – (float) Learning rate
• adam_epsilon – (float) the epsilon value for the adam optimizer
• val_interval – (int) Report training and validation losses every n epochs. By default,
every 10th of the maximum number of epochs.
Returns (BaseRLModel) the pretrained model
1.17. Base RL Class 63

save(save_path, cloudpickle=False)
Save the current parameters to file
Parameters
• save_path – (str or file-like) The save location
• cloudpickle – (bool) Use older cloudpickle format instead of zip-archives.
set_env(env)
Checks the validity of the environment, and if it is coherent, set it as the current environment.
Parameters env – (Gym Environment) The environment for learning a policy
set_random_seed(seed: Optional[int]) → None
Parameters seed – (Optional[int]) Seed for the pseudo-random generators. If None, do not
change the seeds.
setup_model()
Create all the functions and tensorflow graphs necessary to train the model
1.18 Policy Networks
Stable-baselines provides a set of default policies, that can be used with most action spaces. To customize the default
policies, you can specify the policy_kwargs parameter to the model class you use. Those kwargs are then passed
to the policy on instantiation (see Custom Policy Network for an example). If you need more control on the policy
architecture, you can also create a custom policy (see Custom Policy Network).
Note: CnnPolicies are for images only. MlpPolicies are made for other type of features (e.g. robot joints)
Warning: For all algorithms (except DDPG, TD3 and SAC), continuous actions are clipped during training and
testing (to avoid out of bound error).
Available Policies
MlpPolicy Policy object that implements actor critic, using a MLP

(2 layers of 64)
MlpLstmPolicy Policy object that implements actor critic, using LSTMs
with a MLP feature extraction
MlpLnLstmPolicy Policy object that implements actor critic, using a layer
normalized LSTMs with a MLP feature extraction
CnnPolicy Policy object that implements actor critic, using a CNN
(the nature CNN)
CnnLstmPolicy Policy object that implements actor critic, using LSTMs
with a CNN feature extraction
CnnLnLstmPolicy Policy object that implements actor critic, using a layer
normalized LSTMs with a CNN feature extraction

1.18.1 Base Classes
class stable_baselines.common.policies.BasePolicy(sess, ob_space, ac_space, n_env,

n_steps, n_batch, reuse=False,
scale=False, obs_phs=None,
add_action_ph=False)
The base policy object
Parameters
• sess – (TensorFlow session) The current TensorFlow session
• ob_space – (Gym Space) The observation space of the environment
• ac_space – (Gym Space) The action space of the environment
• n_env – (int) The number of environments to run
• n_steps – (int) The number of steps to run for each environment
• n_batch – (int) The number of batches to run (n_envs * n_steps)
• reuse – (bool) If the policy is reusable or not
• scale – (bool) whether or not to scale the input
• obs_phs – (TensorFlow Tensor, TensorFlow Tensor) a tuple containing an override for
observation placeholder and the processed observation placeholder respectively
• add_action_ph – (bool) whether or not to create an action placeholder
action_ph
tf.Tensor: placeholder for actions, shape (self.n_batch, ) + self.ac_space.shape.
initial_state
The initial state of the policy. For feedforward policies, None. For a recurrent policy, a NumPy array of
shape (self.n_env, ) + state_shape.
is_discrete
bool: is action space discrete.
obs_ph
tf.Tensor: placeholder for observations, shape (self.n_batch, ) + self.ob_space.shape.
proba_step(obs, state=None, mask=None)
Returns the action probability for a single step
Parameters
• obs – ([float] or [int]) The current observation of the environment
• state – ([float]) The last states (used in recurrent policies)
• mask – ([float]) The last masks (used in recurrent policies)
Returns ([float]) the action probability
processed_obs
tf.Tensor: processed observations, shape (self.n_batch, ) + self.ob_space.shape.
The form of processing depends on the type of the observation space, and the parameters whether scale is
passed to the constructor; see observation_input for more information.
step(obs, state=None, mask=None)
Returns the policy for a single step
Parameters
1.18. Policy Networks 65


Returns ([float], [float], [float], [float]) actions, values, states, neglogp
class stable_baselines.common.policies.ActorCriticPolicy(sess, ob_space,
ac_space, n_env, n_steps,
n_batch, reuse=False,
scale=False)
Policy object that implements actor critic
Parameters
• n_batch – (int) The number of batch to run (n_envs * n_steps)
• scale – (bool) whether or not to scale the input
action
tf.Tensor: stochastic action, of shape (self.n_batch, ) + self.ac_space.shape.
deterministic_action
tf.Tensor: deterministic action, of shape (self.n_batch, ) + self.ac_space.shape.
neglogp
tf.Tensor: negative log likelihood of the action sampled by self.action.
pdtype
ProbabilityDistributionType: type of the distribution for stochastic actions.
policy
tf.Tensor: policy output, e.g. logits.
policy_proba
tf.Tensor: parameters of the probability distribution. Depends on pdtype.
proba_distribution
ProbabilityDistribution: distribution of stochastic actions.
step(obs, state=None, mask=None, deterministic=False)
Parameters

value(obs, state=None, mask=None)

Returns the value for a single step
Parameters
Returns ([float]) The associated value of the action
value_flat
tf.Tensor: value estimate, of shape (self.n_batch, )
value_fn
tf.Tensor: value estimate, of shape (self.n_batch, 1)
class stable_baselines.common.policies.FeedForwardPolicy(sess, ob_space,
ac_space, n_env, n_steps,
n_batch, reuse=False,
layers=None,
net_arch=None,
act_fun=<MagicMock
id=’140452378410640’>,
cnn_extractor=<function
nature_cnn>, fea-
ture_extraction=’cnn’,
**kwargs)
Policy object that implements actor critic, using a feed forward neural network.
Parameters
• layers – ([int]) (deprecated, use net_arch instead) The size of the Neural network for the
policy (if None, default to [64, 64])
• net_arch – (list) Specification of the actor-critic policy network architecture (see
mlp_extractor documentation for details).
• act_fun – (tf.func) the activation function to use in the neural network.
• cnn_extractor – (function (TensorFlow Tensor, **kwargs): (TensorFlow Tensor))
the CNN feature extraction
• feature_extraction – (str) The feature extraction type (“cnn” or “mlp”)
• kwargs – (dict) Extra keyword arguments for the nature CNN feature extraction

Parameters
Parameters
Parameters
class stable_baselines.common.policies.LstmPolicy(sess, ob_space, ac_space,
n_env, n_steps, n_batch,
n_lstm=256, reuse=False, lay-
ers=None, net_arch=None,
act_fun=<MagicMock
id=’140452378442640’>,
cnn_extractor=<function na-
ture_cnn>, layer_norm=False, fea-
ture_extraction=’cnn’, **kwargs)
Policy object that implements actor critic, using LSTMs.
Parameters
• n_lstm – (int) The number of LSTM cells (for recurrent policies)

• layers – ([int]) The size of the Neural network before the LSTM layer (if None, default
to [64, 64])
• net_arch – (list) Specification of the actor-critic policy network architecture. Notation
similar to the format described in mlp_extractor but with additional support for a ‘lstm’
entry in the shared network part.
• act_fun – (tf.func) the activation function to use in the neural network.
• cnn_extractor – (function (TensorFlow Tensor, **kwargs): (TensorFlow Tensor))
the CNN feature extraction
• layer_norm – (bool) Whether or not to use layer normalizing LSTMs
• feature_extraction – (str) The feature extraction type (“cnn” or “mlp”)
Parameters
Parameters
Cf base class doc.
1.18.2 MLP Policies
class stable_baselines.common.policies.MlpPolicy(sess, ob_space, ac_space, n_env,

**_kwargs)
Policy object that implements actor critic, using a MLP (2 layers of 64)
Parameters


• _kwargs – (dict) Extra keyword arguments for the nature CNN feature extraction
class stable_baselines.common.policies.MlpLstmPolicy(sess, ob_space, ac_space,
n_lstm=256, reuse=False,
**_kwargs)
Policy object that implements actor critic, using LSTMs with a MLP feature extraction
Parameters
class stable_baselines.common.policies.MlpLnLstmPolicy(sess, ob_space, ac_space,
**_kwargs)
Policy object that implements actor critic, using a layer normalized LSTMs with a MLP feature extraction
Parameters
1.18.3 CNN Policies
class stable_baselines.common.policies.CnnPolicy(sess, ob_space, ac_space, n_env,

**_kwargs)
Policy object that implements actor critic, using a CNN (the nature CNN)

Parameters
class stable_baselines.common.policies.CnnLstmPolicy(sess, ob_space, ac_space,
**_kwargs)
Policy object that implements actor critic, using LSTMs with a CNN feature extraction
Parameters
class stable_baselines.common.policies.CnnLnLstmPolicy(sess, ob_space, ac_space,
**_kwargs)
Policy object that implements actor critic, using a layer normalized LSTMs with a CNN feature extraction
Parameters

1.19 A2C
A synchronous, deterministic variant of Asynchronous Advantage Actor Critic (A3C). It uses multiple workers to
avoid the use of a replay buffer.
1.19.1 Notes
• Original paper: https://arxiv.org/abs/1602.01783

• OpenAI blog post: https://openai.com/blog/baselines-acktr-a2c/
• python -m stable_baselines.a2c.run_atari runs the algorithm for 40M frames = 10M
timesteps on an Atari game. See help (-h) for more options.
• python -m stable_baselines.a2c.run_mujoco runs the algorithm for 1M frames on a Mujoco
environment.
1.19.2 Can I use?
• Recurrent policies: X
• Multi processing: X
• Gym spaces:
Space Action Observation

Discrete X X
Box X X
MultiDiscrete X X
MultiBinary X X
1.19.3 Example
Train a A2C agent on CartPole-v1 using 4 processes.
import gym

from stable_baselines.common import make_vec_env
# Parallel environments
env = make_vec_env('CartPole-v1', n_envs=4)
model = A2C(MlpPolicy, env, verbose=1)

model.save("a2c_cartpole")
del model # remove to demonstrate saving and loading
model = A2C.load("a2c_cartpole")
obs = env.reset()


while True:
env.render()
1.19.4 Parameters
class stable_baselines.a2c.A2C(policy, env, gamma=0.99, n_steps=5, vf_coef=0.25,

ent_coef=0.01, max_grad_norm=0.5, learning_rate=0.0007,
alpha=0.99, momentum=0.0, epsilon=1e-05,
lr_schedule=’constant’, verbose=0, tensorboard_log=None,
_init_setup_model=True, policy_kwargs=None,
full_tensorboard_log=False, seed=None, n_cpu_tf_sess=None)
The A2C (Advantage Actor Critic) model class, https://arxiv.org/abs/1602.01783
Parameters
• policy – (ActorCriticPolicy or str) The policy model to use (MlpPolicy, CnnPolicy, CnnL-
stmPolicy, . . . )
• env – (Gym environment or str) The environment to learn from (if registered in Gym, can
be str)
• gamma – (float) Discount factor
• n_steps – (int) The number of steps to run for each environment per update (i.e. batch
size is n_steps * n_env where n_env is number of environment copies running in parallel)
• vf_coef – (float) Value function coefficient for the loss calculation
• ent_coef – (float) Entropy coefficient for the loss calculation
• max_grad_norm – (float) The maximum value for the gradient clipping
• learning_rate – (float) The learning rate
• alpha – (float) RMSProp decay parameter (default: 0.99)
• momentum – (float) RMSProp momentum parameter (default: 0.0)
• epsilon – (float) RMSProp epsilon (stabilizes square root computation in denominator of
RMSProp update) (default: 1e-5)
• lr_schedule – (str) The type of scheduler for the learning rate update (‘linear’, ‘con-
stant’, ‘double_linear_con’, ‘middle_drop’ or ‘double_middle_drop’)
• tensorboard_log – (str) the log location for tensorboard (if None, no logging)
• _init_setup_model – (bool) Whether or not to build the network at the creation of the
instance (used only for loading)
• full_tensorboard_log – (bool) enable additional logging when using tensorboard
WARNING: this logging can take a lot of space quickly
1.19. A2C 73
explanation
Parameters
distribution)
get_env()
get_parameters()
learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’A2C’, re-
Parameters


more details.
(used in logging)
Parameters
Parameters
Defaults to True.

Parameters
1.19. A2C 75
cies)
Parameters
Parameters
set_env(env)
change the seeds.
setup_model()
1.19.5 Callbacks - Accessible Variables
Depending on initialization parameters and timestep, different variables are accessible. Variables accessible “From
timestep X” are variables that can be accessed when self.timestep==X in the on_step function.

Variable Availability
From timestep 1
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• reset_num_timesteps
• new_tb_log
• writer
• t_start
• mb_obs
• mb_rewards
• mb_actions
• mb_values
• mb_dones
• mb_states
• ep_infos
• actions
• values
• states
• clipped_actions
• obs
• rewards
• dones
• infos
From timestep 2
• info
• maybe_ep_info
From timestep n_step+1

• update
• rollout
• masks
• true_reward
From timestep 2 * n_step+1

• value_loss
• policy_entropy
• n_seconds
• fps
1.20 ACER
Sample Efficient Actor-Critic with Experience Replay (ACER) combines several ideas of previous al-
gorithms: it uses multiple workers (as A2C), implements a replay buffer (as in DQN), uses Retrace for
Q-value estimation, importance sampling and a trust region.
1.20. ACER 77
1.20.1 Notes

• python -m stable_baselines.acer.run_atari runs the algorithm for 40M frames = 10M
1.20.2 Can I use?
• Gym spaces:

Discrete X X
Box X
MultiDiscrete X
MultiBinary X
1.20.3 Example
import gym
from stable_baselines.common.policies import MlpPolicy, MlpLstmPolicy, MlpLnLstmPolicy

from stable_baselines import ACER
# multiprocess environment
model = ACER(MlpPolicy, env, verbose=1)

model.save("acer_cartpole")
model = ACER.load("acer_cartpole")
obs = env.reset()
while True:
env.render()

1.20.4 Parameters
class stable_baselines.acer.ACER(policy, env, gamma=0.99, n_steps=20, num_procs=None,

q_coef=0.5, ent_coef=0.01, max_grad_norm=10, learn-
ing_rate=0.0007, lr_schedule=’linear’, rprop_alpha=0.99,
rprop_epsilon=1e-05, buffer_size=5000, replay_ratio=4, re-
play_start=1000, correction_term=10.0, trust_region=True,
alpha=0.99, delta=1, verbose=0, tensorboard_log=None,
full_tensorboard_log=False, seed=None, n_cpu_tf_sess=1)
The ACER (Actor-Critic with Experience Replay) model class, https://arxiv.org/abs/1611.01224
Parameters
stmPolicy, . . . )
be str)
• gamma – (float) The discount value
• num_procs – (int) The number of threads for TensorFlow operations
Deprecated since version 2.9.0: Use n_cpu_tf_sess instead.
• q_coef – (float) The weight for the loss on the Q value
• ent_coef – (float) The weight for the entropy loss
• max_grad_norm – (float) The clipping value for the maximum gradient
• learning_rate – (float) The initial learning rate for the RMS prop optimizer
• rprop_epsilon – (float) RMSProp epsilon (stabilizes square root computation in de-
nominator of RMSProp update) (default: 1e-5)
• rprop_alpha – (float) RMSProp decay parameter (default: 0.99)
• buffer_size – (int) The buffer size in number of steps
• replay_ratio – (float) The number of replay learning per on policy learning on average,
using a poisson distribution
• replay_start – (int) The minimum number of steps in the buffer, before learning replay
• correction_term – (float) Importance weight clipping factor (default: 10)
• trust_region – (bool) Whether or not algorithms estimates the gradient KL divergence
between the old and updated policy and uses it to determine step size (default: True)
• alpha – (float) The decay rate for the Exponential moving average of the parameters
• delta – (float) max KL divergence between the old policy and updated policy (default: 1)
1.20. ACER 79
instance
explanation
Parameters
distribution)
get_env()
get_parameters()

learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’ACER’, re-
Parameters
more details.
(used in logging)
Parameters
Parameters
1.20. ACER 81
Defaults to True.

Parameters
cies)
Parameters
Parameters
set_env(env)
change the seeds.
setup_model()

Depending on initialization parameters and timestep, different variables are accessible. Variables accessible from
“timestep X” are variables that can be accessed when self.timestep==X from the on_step function.
From timestep 1
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• new_tb_log
• writer
• episode_stats
• buffer
• t_start
• enc_obs
• mb_obs
• mb_actions
• mb_mus
• mb_dones
• mb_rewards
• actions
• states
• mus
• clipped_actions
• obs
• rewards
• dones
From timestep n_step+1

• steps
• masks
• names_ops From timestep 2 * n_step+1

• values_ops
After replay_start steps, when replay_ratio > 0

• samples_number
and buffer is not None
1.21 ACKTR
Actor Critic using Kronecker-Factored Trust Region (ACKTR) uses Kronecker-factored approximate curvature (K-
FAC) for trust region optimization.
1.21. ACKTR 83
1.21.1 Notes

• Baselines blog post: https://blog.openai.com/baselines-acktr-a2c/
• python -m stable_baselines.acktr.run_atari runs the algorithm for 40M frames = 10M
1.21.2 Can I use?
• Gym spaces:

Discrete X X
Box X X
MultiDiscrete X
MultiBinary X
1.21.3 Example
import gym
from stable_baselines.common.policies import MlpPolicy, MlpLstmPolicy, MlpLnLstmPolicy

from stable_baselines import ACKTR
model = ACKTR(MlpPolicy, env, verbose=1)

model.save("acktr_cartpole")
model = ACKTR.load("acktr_cartpole")
obs = env.reset()
while True:
env.render()

1.21.4 Parameters
class stable_baselines.acktr.ACKTR(policy, env, gamma=0.99, nprocs=None, n_steps=20,

ent_coef=0.01, vf_coef=0.25, vf_fisher_coef=1.0, learn-
ing_rate=0.25, max_grad_norm=0.5, kfac_clip=0.001,
lr_schedule=’linear’, verbose=0, tensorboard_log=None,
_init_setup_model=True, async_eigen_decomp=False,
kfac_update=1, gae_lambda=None, pol-
icy_kwargs=None, full_tensorboard_log=False,
seed=None, n_cpu_tf_sess=1)
The ACKTR (Actor Critic using Kronecker-Factored Trust Region) model class, https://arxiv.org/abs/1708.
05144
Parameters
stmPolicy, . . . )
be str)
• nprocs – (int) The number of threads for TensorFlow operations
Deprecated since version 2.9.0: Use n_cpu_tf_sess instead.
• ent_coef – (float) The weight for the entropy loss
• vf_coef – (float) The weight for the loss on the value function
• vf_fisher_coef – (float) The weight for the fisher loss on the value function
• learning_rate – (float) The initial learning rate for the RMS prop optimizer
• max_grad_norm – (float) The clipping value for the maximum gradient
• kfac_clip – (float) gradient clipping for Kullback-Leibler
instance
• async_eigen_decomp – (bool) Use async eigen decomposition
• kfac_update – (int) update kfac after kfac_update steps
• gae_lambda – (float) Factor for trade-off of bias vs variance for Generalized Advantage
Estimator If None (default), then the classic advantage will be used instead of GAE
1.21. ACKTR 85
explanation
Parameters
distribution)
get_env()
get_parameters()
learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’ACKTR’, re-
Parameters


more details.
(used in logging)
Parameters
Parameters
Defaults to True.

Parameters
1.21. ACKTR 87
cies)
Parameters
Parameters
set_env(env)
change the seeds.
setup_model()

From timestep 1
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• new_tb_log
• writer
• tf_vars
• is_uninitialized
• new_uninitialized_vars
• t_start
• coord
• enqueue_threads
• old_uninitialized_vars
• mb_obs
• mb_rewards
• mb_actions
• mb_values
• mb_dones
• mb_states
• ep_infos
• _
• actions
• values
• states
• clipped_actions
• obs
• rewards
• dones
• infos
From timestep 2
• info
• maybe_ep_info
From timestep n_steps+1

• update
• rollout
• returns
• masks
• true_reward
From timestep 2*n_steps+1

• policy_loss
• value_loss
• policy_entropy
• n_seconds
• fps
1.21. ACKTR 89
1.22 DDPG
Deep Deterministic Policy Gradient (DDPG)
Note: DDPG requires OpenMPI. If OpenMPI isn’t enabled, then DDPG isn’t imported into the
stable_baselines module.
Warning: The DDPG model does not support stable_baselines.common.policies because it uses
q-value instead of value estimation, as a result it must use its own policy models (see DDPG Policies).
Available Policies

(2 layers of 64)
LnMlpPolicy Policy object that implements actor critic, using a MLP
(2 layers of 64), with layer normalisation
(the nature CNN)
LnCnnPolicy Policy object that implements actor critic, using a CNN
(the nature CNN), with layer normalisation
1.22.1 Notes

• Baselines post: https://blog.openai.com/better-exploration-with-parameter-noise/
• python -m stable_baselines.ddpg.main runs the algorithm for 1M frames = 10M timesteps on a
Mujoco environment. See help (-h) for more options.
1.22.2 Can I use?
• Recurrent policies:
• Multi processing: X (using MPI)
• Gym spaces:

Discrete X
Box X X
MultiDiscrete X
MultiBinary X
1.22.3 Example

import gym
import numpy as np
from stable_baselines.ddpg.policies import MlpPolicy

from stable_baselines.common.noise import NormalActionNoise,
˓→OrnsteinUhlenbeckActionNoise, AdaptiveParamNoiseSpec
env = gym.make('MountainCarContinuous-v0')
# the noise objects for DDPG

n_actions = env.action_space.shape[-1]
param_noise = None
action_noise = OrnsteinUhlenbeckActionNoise(mean=np.zeros(n_actions), sigma=float(0.
˓→5) * np.ones(n_actions))
model = DDPG(MlpPolicy, env, verbose=1, param_noise=param_noise, action_noise=action_

˓→noise)
model.save("ddpg_mountain")
model = DDPG.load("ddpg_mountain")
obs = env.reset()
while True:
env.render()
1.22.4 Parameters
class stable_baselines.ddpg.DDPG(policy, env, gamma=0.99, memory_policy=None,

eval_env=None, nb_train_steps=50,
nb_rollout_steps=100, nb_eval_steps=100,
param_noise=None, action_noise=None, normal-
ize_observations=False, tau=0.001, batch_size=128,
param_noise_adaption_interval=50, nor-
malize_returns=False, enable_popart=False,
observation_range=(-5.0, 5.0), critic_l2_reg=0.0,
return_range=(-inf, inf), actor_lr=0.0001, critic_lr=0.001,
clip_norm=None, reward_scale=1.0, render=False, ren-
der_eval=False, memory_limit=None, buffer_size=50000,
random_exploration=0.0, verbose=0, tensor-
board_log=None, _init_setup_model=True, pol-
icy_kwargs=None, full_tensorboard_log=False, seed=None,
n_cpu_tf_sess=1)
Deep Deterministic Policy Gradient (DDPG) model
DDPG: https://arxiv.org/pdf/1509.02971.pdf
Parameters
• policy – (DDPGPolicy or str) The policy model to use (MlpPolicy, CnnPolicy, LnMlp-
Policy, . . . )
1.22. DDPG 91
be str)
• gamma – (float) the discount factor
• memory_policy – (ReplayBuffer) the replay buffer (if None, default to base-
lines.deepq.replay_buffer.ReplayBuffer)
Deprecated since version 2.6.0: This parameter will be removed in a future version
• eval_env – (Gym Environment) the evaluation environment (can be None)
• nb_train_steps – (int) the number of training steps
• nb_rollout_steps – (int) the number of rollout steps
• nb_eval_steps – (int) the number of evaluation steps
• param_noise – (AdaptiveParamNoiseSpec) the parameter noise type (can be None)
• action_noise – (ActionNoise) the action noise type (can be None)
• param_noise_adaption_interval – (int) apply param noise every N steps
• tau – (float) the soft update coefficient (keep old values, between 0 and 1)
• normalize_returns – (bool) should the critic output be normalized
• enable_popart – (bool) enable pop-art normalization of the critic output (https://arxiv.
org/pdf/1602.07714.pdf), normalize_returns must be set to True.
• normalize_observations – (bool) should the observation be normalized
• batch_size – (int) the size of the batch for learning the policy
• observation_range – (tuple) the bounding values for the observation
• return_range – (tuple) the bounding values for the critic output
• critic_l2_reg – (float) l2 regularizer coefficient
• actor_lr – (float) the actor learning rate
• critic_lr – (float) the critic learning rate
• clip_norm – (float) clip the gradients (disabled if None)
• reward_scale – (float) the value the reward should be scaled by
• render – (bool) enable rendering of the environment
• render_eval – (bool) enable rendering of the evaluation environment
• memory_limit – (int) the max number of transitions to store, size of the replay buffer
Deprecated since version 2.6.0: Use buffer_size instead.
• buffer_size – (int) the max number of transitions to store, size of the replay buffer
• random_exploration – (float) Probability of taking a random action (as in an epsilon-
greedy strategy) This is not needed for DDPG normally but can help exploring when using
HER + DDPG. This hack was present in the original OpenAI Baselines repo (DDPG + HER)
instance


explanation
Parameters
distribution)
get_env()
get_parameters()
1.22. DDPG 93
is_using_her() → bool
Check if is using HER
Returns (bool) Whether is using HER or not
learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’DDPG’, re-
set_num_timesteps=True, replay_wrapper=None)
Parameters
more details.
(used in logging)
Parameters
Parameters

Defaults to True.
predict(observation, state=None, mask=None, deterministic=True)

Parameters
cies)
Parameters
replay_buffer_add(obs_t, action, reward, obs_tp1, done, info)
Add a new transition to the replay buffer
Parameters
• obs_t – (np.ndarray) the last observation
• action – ([float]) the action
• reward – (float) the reward of the transition
• obs_tp1 – (np.ndarray) the new observation
• done – (bool) is the episode done
• info – (dict) extra values used to compute the reward when using HER
Parameters
set_env(env)
1.22. DDPG 95

change the seeds.
setup_model()
1.22.5 DDPG Policies
class stable_baselines.ddpg.MlpPolicy(sess, ob_space, ac_space, n_env, n_steps, n_batch,

reuse=False, **_kwargs)
Parameters
action_ph
initial_state
is_discrete
make_actor(obs=None, reuse=False, scope=’pi’)
creates an actor object
Parameters
• obs – (TensorFlow Tensor) The observation placeholder (can be None for default place-
holder)
• reuse – (bool) whether or not to reuse parameters
• scope – (str) the scope name of the actor
Returns (TensorFlow Tensor) the output tensor
make_critic(obs=None, action=None, reuse=False, scope=’qf’)
creates a critic object
Parameters
holder)

• action – (TensorFlow Tensor) The action placeholder (can be None for default place-
holder)
• scope – (str) the scope name of the critic
obs_ph
Parameters
processed_obs
Parameters
Returns ([float]) actions
value(obs, action, state=None, mask=None)
Parameters
• action – ([float] or [int]) The taken action
class stable_baselines.ddpg.LnMlpPolicy(sess, ob_space, ac_space, n_env, n_steps, n_batch,
Policy object that implements actor critic, using a MLP (2 layers of 64), with layer normalisation
Parameters
1.22. DDPG 97

action_ph
initial_state
is_discrete
Parameters
holder)
Parameters
holder)
holder)
obs_ph
Parameters

processed_obs
Parameters
Parameters
class stable_baselines.ddpg.CnnPolicy(sess, ob_space, ac_space, n_env, n_steps, n_batch,
Parameters
action_ph
initial_state
is_discrete
1.22. DDPG 99
Parameters
holder)
Parameters
holder)
holder)
obs_ph
Parameters
processed_obs
Parameters
Parameters


class stable_baselines.ddpg.LnCnnPolicy(sess, ob_space, ac_space, n_env, n_steps, n_batch,
Policy object that implements actor critic, using a CNN (the nature CNN), with layer normalisation
Parameters
action_ph
initial_state
is_discrete
Parameters
holder)
Parameters
holder)
holder)
1.22. DDPG 101


obs_ph
Parameters
processed_obs
Parameters
Parameters
1.22.6 Action and Parameters Noise
class stable_baselines.ddpg.AdaptiveParamNoiseSpec(initial_stddev=0.1, de-

sired_action_stddev=0.1, adop-
tion_coefficient=1.01)
Implements adaptive parameter noise
Parameters
• initial_stddev – (float) the initial value for the standard deviation of the noise

• desired_action_stddev – (float) the desired value for the standard deviation of the
noise
• adoption_coefficient – (float) the update coefficient for the standard deviation of
the noise
adapt(distance)
update the standard deviation for the parameter noise
Parameters distance – (float) the noise distance applied to the parameters
get_stats()
return the standard deviation for the parameter noise
Returns (dict) the stats of the noise
class stable_baselines.ddpg.NormalActionNoise(mean, sigma)
A Gaussian action noise
Parameters
• mean – (float) the mean value of the noise
• sigma – (float) the scale of the noise (std here)
reset() → None
call end of episode reset for the noise
class stable_baselines.ddpg.OrnsteinUhlenbeckActionNoise(mean, sigma,
theta=0.15, dt=0.01,
initial_noise=None)
A Ornstein Uhlenbeck action noise, this is designed to approximate brownian motion with friction.
Based on http://math.stackexchange.com/questions/1287634/implementing-ornstein-uhlenbeck-in-matlab
Parameters
• mean – (float) the mean of the noise
• sigma – (float) the scale of the noise
• theta – (float) the rate of mean reversion
• dt – (float) the timestep for the noise
• initial_noise – ([float]) the initial value for the noise output, (if None: 0)
reset() → None
reset the Ornstein Uhlenbeck noise, to the initial position
Similarly to the example given in the examples page. You can easily define a custom architecture for the policy
network:
import gym
from stable_baselines.ddpg.policies import FeedForwardPolicy

# Custom MLP policy of two layers of size 16 each

class CustomDDPGPolicy(FeedForwardPolicy):
1.22. DDPG 103


super(CustomDDPGPolicy, self).__init__(*args, **kwargs,
layers=[16, 16],
layer_norm=False,
model = DDPG(CustomDDPGPolicy, 'Pendulum-v0', verbose=1)

# Train the agent

From timestep 1
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• replay_wrapper
• new_tb_log
• writer
• rank
• eval_episode_rewards_history
• episode_rewards_history
• episode_successes
• obs
• eval_obs
• episode_reward
• episode_step
• episodes
• step
• total_steps
• start_time
• epoch_episode_rewards
• epoch_episode_steps
• epoch_actor_losses
• epoch_critic_losses
• epoch_adaptive_distances
• eval_episode_rewards
• eval_qs
• epoch_actions
• epoch_qs
• epoch_episodes
• epoch
• action
• q_value
• unscaled_action
• new_obs
• reward
• done
• info
From timestep 2
• obs_
• new_obs_
• reward_
After nb_rollout_steps+1
• t_train
After nb_rollout_steps*ceil(nb_rollout_steps/batch_size)‘‘‘
• distance
• critic_loss
• actor_loss
After episode termination

• maybe_is_success
1.22. DDPG 105
1.23 DQN
Deep Q Network (DQN) and its extensions (Double-DQN, Dueling-DQN, Prioritized Experience Replay).
Warning: The DQN model does not support stable_baselines.common.policies, as a result it must
use its own policy models (see DQN Policies).
Available Policies
MlpPolicy Policy object that implements DQN policy, using a MLP

(2 layers of 64)
LnMlpPolicy Policy object that implements DQN policy, using a MLP
CnnPolicy Policy object that implements DQN policy, using a
CNN (the nature CNN)
LnCnnPolicy Policy object that implements DQN policy, using a
CNN (the nature CNN), with layer normalisation
1.23.1 Notes
• DQN paper: https://arxiv.org/abs/1312.5602

• Dueling DQN: https://arxiv.org/abs/1511.06581
• Double-Q Learning: https://arxiv.org/abs/1509.06461
• Prioritized Experience Replay: https://arxiv.org/abs/1511.05952
Note: By default, the DQN class has double q learning and dueling extensions enabled. See Issue #406 for disabling
dueling. To disable double-q learning, you can change the default value in the constructor.
1.23.2 Can I use?
• Multi processing:
• Gym spaces:

Discrete X X
Box X
MultiDiscrete X
MultiBinary X

1.23.3 Example
import gym

from stable_baselines.deepq.policies import MlpPolicy
model = DQN(MlpPolicy, env, verbose=1)

model.save("deepq_cartpole")
model = DQN.load("deepq_cartpole")
obs = env.reset()
while True:
env.render()
With Atari:
from stable_baselines.common.atari_wrappers import make_atari

from stable_baselines.deepq.policies import MlpPolicy, CnnPolicy
env = make_atari('BreakoutNoFrameskip-v4')
model = DQN(CnnPolicy, env, verbose=1)

model.save("deepq_breakout")
model = DQN.load("deepq_breakout")
obs = env.reset()
while True:
env.render()
1.23. DQN 107

1.23.4 Parameters
class stable_baselines.deepq.DQN(policy, env, gamma=0.99, learning_rate=0.0005,

buffer_size=50000, exploration_fraction=0.1, explo-
ration_final_eps=0.02, exploration_initial_eps=1.0,
train_freq=1, batch_size=32, double_q=True, learn-
ing_starts=1000, target_network_update_freq=500, pri-
oritized_replay=False, prioritized_replay_alpha=0.6,
prioritized_replay_beta0=0.4, priori-
tized_replay_beta_iters=None, prioritized_replay_eps=1e-
06, param_noise=False, n_cpu_tf_sess=None, verbose=0,
tensorboard_log=None, _init_setup_model=True, pol-
icy_kwargs=None, full_tensorboard_log=False, seed=None)
The DQN model class. DQN paper: https://arxiv.org/abs/1312.5602 Dueling DQN: https://arxiv.org/abs/1511.
06581 Double-Q Learning: https://arxiv.org/abs/1509.06461 Prioritized Experience Replay: https://arxiv.org/
abs/1511.05952
Parameters
• policy – (DQNPolicy or str) The policy model to use (MlpPolicy, CnnPolicy, LnMlpPol-
icy, . . . )
be str)
• learning_rate – (float) learning rate for adam optimizer
• buffer_size – (int) size of the replay buffer
• exploration_fraction – (float) fraction of entire training period over which the ex-
ploration rate is annealed
• exploration_final_eps – (float) final value of random action probability
• exploration_initial_eps – (float) initial value of random action probability
• train_freq – (int) update the model every train_freq steps. set to None to disable print-
ing
• batch_size – (int) size of a batched sampled from replay buffer for training
• double_q – (bool) Whether to enable Double-Q learning or not.
• learning_starts – (int) how many steps of the model to collect transitions for before
learning starts
• target_network_update_freq – (int) update the target network every tar-
get_network_update_freq steps.
• prioritized_replay – (bool) if True prioritized replay buffer will be used.
• prioritized_replay_alpha – (float)alpha parameter for prioritized replay buffer.
It determines how much prioritization is used, with alpha=0 corresponding to the uniform
case.
• prioritized_replay_beta0 – (float) initial value of beta for prioritized replay buffer
• prioritized_replay_beta_iters – (int) number of iterations over which beta will
be annealed from initial value to 1.0. If set to None equals to max_timesteps.
• prioritized_replay_eps – (float) epsilon to add to the TD errors when updating
priorities.

• param_noise – (bool) Whether or not to apply noise to the parameters of the policy.
instance
explanation
Parameters
distribution)
get_env()
get_parameters()
1.23. DQN 109


learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’DQN’, re-
Parameters
more details.
(used in logging)
Parameters

Parameters
Defaults to True.

Parameters
cies)
Parameters
Parameters
Parameters
1.23. DQN 111


set_env(env)
change the seeds.
setup_model()
1.23.5 DQN Policies
class stable_baselines.deepq.MlpPolicy(sess, ob_space, ac_space, n_env, n_steps, n_batch,

reuse=False, obs_phs=None, dueling=True,
**_kwargs)
Policy object that implements DQN policy, using a MLP (2 layers of 64)
Parameters
• dueling – (bool) if true double the output MLP to compute a baseline for action scores
action_ph
initial_state
is_discrete
obs_ph
Parameters
• obs – (np.ndarray float or int) The current observation of the environment

• state – (np.ndarray float) The last states (used in recurrent policies)

• mask – (np.ndarray float) The last masks (used in recurrent policies)
Returns (np.ndarray float) the action probability
processed_obs
step(obs, state=None, mask=None, deterministic=True)
Returns the q_values for a single step
Parameters
Returns (np.ndarray int, np.ndarray float, np.ndarray float) actions, q_values, states
class stable_baselines.deepq.LnMlpPolicy(sess, ob_space, ac_space, n_env, n_steps,
n_batch, reuse=False, obs_phs=None, duel-
ing=True, **_kwargs)
Policy object that implements DQN policy, using a MLP (2 layers of 64), with layer normalisation
Parameters
action_ph
initial_state
is_discrete
obs_ph
1.23. DQN 113


Parameters
processed_obs
Parameters
class stable_baselines.deepq.CnnPolicy(sess, ob_space, ac_space, n_env, n_steps, n_batch,
reuse=False, obs_phs=None, dueling=True,
**_kwargs)
Policy object that implements DQN policy, using a CNN (the nature CNN)
Parameters
action_ph
initial_state

is_discrete
obs_ph
Parameters
processed_obs
Parameters
class stable_baselines.deepq.LnCnnPolicy(sess, ob_space, ac_space, n_env, n_steps,
n_batch, reuse=False, obs_phs=None, duel-
ing=True, **_kwargs)
Policy object that implements DQN policy, using a CNN (the nature CNN), with layer normalisation
Parameters
1.23. DQN 115

action_ph
initial_state
is_discrete
obs_ph
Parameters
processed_obs
Parameters
network:
import gym
from stable_baselines.deepq.policies import FeedForwardPolicy

# Custom MLP policy of two layers of size 32 each

class CustomDQNPolicy(FeedForwardPolicy):
super(CustomDQNPolicy, self).__init__(*args, **kwargs,


layers=[32, 32],
layer_norm=False,

model = DQN(CustomDQNPolicy, env, verbose=1)

# Train the agent
1.23. DQN 117

From timestep 1
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• replay_wrapper
• new_tb_log
• writer
• episode_rewards
• reset
• obs
• _
• kwargs
• update_eps
• update_param_noise_threshold
• action
• env_action
• new_obs
• rew
• done
• info
From timestep 2
• obs_
• new_obs_
• reward_
• can_sample
• mean_100ep_reward
• num_episodes
After the first episode

After at least max(batch_size,

• obses_t
learning_starts) and every train_freq
• actions
steps
• rewards
• obses_tp1
• dones
• weights
• batch_idxes
• td_errors
1.24 GAIL
The Generative Adversarial Imitation Learning (GAIL) uses expert trajectories to recover a cost function and then
learn a policy.

Learning a cost function from expert demonstrations is called Inverse Reinforcement Learning (IRL). The connection
between GAIL and Generative Adversarial Networks (GANs) is that it uses a discriminator that tries to separate expert
trajectory from trajectories of the learned policy, which has the role of the generator here.
Note: GAIL requires OpenMPI. If OpenMPI isn’t enabled, then GAIL isn’t imported into the stable_baselines
module.
1.24.1 Notes
Warning: Images are not yet handled properly by the current implementation
1.24.2 If you want to train an imitation learning agent
Step 1: Generate expert data
You can either train a RL algorithm in a classic setting, use another controller (e.g. a PID controller) or human
demonstrations.
We recommend you to take a look at pre-training section or directly look at stable_baselines/gail/
dataset/ folder to learn more about the expected format for the dataset.
Here is an example of training a Soft Actor-Critic model to generate expert trajectories for GAIL:

# Generate expert trajectories (train expert)

# Train for 60000 timesteps and record 10 trajectories
# all the data will be saved in 'expert_pendulum.npz' file
generate_expert_traj(model, 'expert_pendulum', n_timesteps=60000, n_episodes=10)
Step 2: Run GAIL
In case you want to run Behavior Cloning (BC)

Use the .pretrain() method (cf guide).
Others
Thanks to the open source:
• @openai/imitation
• @carpedm20/deep-rl-tensorflow
1.24. GAIL 119

1.24.3 Can I use?
• Gym spaces:

Discrete X X
Box X X
MultiDiscrete X
MultiBinary X
1.24.4 Example
import gym
from stable_baselines import GAIL, SAC

from stable_baselines.gail import ExpertDataset, generate_expert_traj
# Generate expert trajectories (train expert)

generate_expert_traj(model, 'expert_pendulum', n_timesteps=100, n_episodes=10)
# Load the expert dataset

dataset = ExpertDataset(expert_path='expert_pendulum.npz', traj_limitation=10,
˓→verbose=1)
model = GAIL('MlpPolicy', 'Pendulum-v0', dataset, verbose=1)

# Note: in practice, you need to train for 1M steps to have a working policy
model.save("gail_pendulum")
model = GAIL.load("gail_pendulum")
env = gym.make('Pendulum-v0')
obs = env.reset()
while True:
env.render()
1.24.5 Parameters
class stable_baselines.gail.GAIL(policy, env, expert_dataset=None, hid-

den_size_adversary=100, adversary_entcoeff=0.001,
g_step=3, d_step=1, d_stepsize=0.0003, verbose=0,
_init_setup_model=True, **kwargs)
Generative Adversarial Imitation Learning (GAIL)

Warning: Images are not yet handled properly by the current implementation
Parameters
stmPolicy, . . . )
be str)
• expert_dataset – (ExpertDataset) the dataset manager
• gamma – (float) the discount value
• timesteps_per_batch – (int) the number of timesteps to run per batch (horizon)
• max_kl – (float) the Kullback-Leibler loss threshold
• cg_iters – (int) the number of iterations for the conjugate gradient calculation
• lam – (float) GAE factor
• entcoeff – (float) the weight for the entropy loss
• cg_damping – (float) the compute gradient dampening factor
• vf_stepsize – (float) the value function stepsize
• vf_iters – (int) the value function’s number iterations for learning
• hidden_size – ([int]) the hidden dimension for the MLP
• g_step – (int) number of steps to train policy in each epoch
• d_step – (int) number of steps to train discriminator in each epoch
• d_stepsize – (float) the reward giver stepsize
instance

explanation
Parameters
1.24. GAIL 121

distribution)
get_env()
get_parameters()
learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’GAIL’, re-
Parameters
more details.
(used in logging)
Parameters

Parameters
Defaults to True.

Parameters
cies)
Parameters
1.24. GAIL 123

Parameters
set_env(env)
change the seeds.
setup_model()
1.25 HER
Hindsight Experience Replay (HER)

HER is a method wrapper that works with Off policy methods (DQN, SAC, TD3 and DDPG for example).
Note: HER was re-implemented from scratch in Stable-Baselines compared to the original OpenAI baselines. If you
want to reproduce results from the paper, please use the rl baselines zoo in order to have the correct hyperparameters
and at least 8 MPI workers with DDPG.
Warning: HER requires the environment to inherits from gym.GoalEnv
Warning: you must pass an environment or wrap it with HERGoalEnvWrapper in order to use the predict
method
1.25.1 Notes

• OpenAI paper: Plappert et al. (2018)
• OpenAI blog post: https://openai.com/blog/ingredients-for-robotics-research/
1.25.2 Can I use?
Please refer to the wrapped model (DQN, SAC, TD3 or DDPG) for that section.

1.25.3 Example
from stable_baselines import HER, DQN, SAC, DDPG, TD3

from stable_baselines.her import GoalSelectionStrategy, HERGoalEnvWrapper
from stable_baselines.common.bit_flipping_env import BitFlippingEnv
model_class = DQN # works also with SAC, DDPG and TD3
env = BitFlippingEnv(N_BITS, continuous=model_class in [DDPG, SAC, TD3], max_steps=N_

˓→BITS)
# Available strategies (cf paper): future, final, episode, random

goal_selection_strategy = 'future' # equivalent to GoalSelectionStrategy.FUTURE
# Wrap the model

model = HER('MlpPolicy', env, model_class, n_sampled_goal=4, goal_selection_
˓→strategy=goal_selection_strategy,
verbose=1)
# Train the model
model.learn(1000)
model.save("./her_bit_env")
# WARNING: you must pass an env

# or wrap your environment with HERGoalEnvWrapper to use the predict method
model = HER.load('./her_bit_env', env=env)
obs = env.reset()
obs, reward, done, _ = env.step(action)
if done:
obs = env.reset()
1.25.4 Parameters
class stable_baselines.her.HER(policy, env, model_class, n_sampled_goal=4,

goal_selection_strategy=’future’, *args, **kwargs)
Hindsight Experience Replay (HER) https://arxiv.org/abs/1707.01495
Parameters
• policy – (BasePolicy or str) The policy model to use (MlpPolicy, CnnPolicy, CnnLstm-
Policy, . . . )
be str)
• model_class – (OffPolicyRLModel) The off policy RL model to apply Hindsight Expe-
rience Replay currently supported: DQN, DDPG, SAC
• n_sampled_goal – (int)
• goal_selection_strategy – (GoalSelectionStrategy or str)
1.25. HER 125


explanation
Parameters
distribution)
get_env()
learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’HER’, re-
Parameters
more details.
(used in logging)


Parameters
Parameters
cies)
Parameters
set_env(env)
setup_model()
1.25.5 Goal Selection Strategies
class stable_baselines.her.GoalSelectionStrategy
The strategies for selecting new goals when creating artificial transitions.
1.25.6 Goal Env Wrapper
class stable_baselines.her.HERGoalEnvWrapper(env)
A wrapper that allow to use dict observation space (coming from GoalEnv) with the RL algorithms. It assumes
that all the spaces of the dict space are of the same type.
Parameters env – (gym.GoalEnv)
1.25. HER 127

convert_dict_to_obs(obs_dict)
Parameters obs_dict – (dict<np.ndarray>)
Returns (np.ndarray)
convert_obs_to_dict(observations)
Inverse operation of convert_dict_to_obs
Parameters observations – (np.ndarray)
Returns (OrderedDict<np.ndarray>)
1.25.7 Replay Wrapper
class stable_baselines.her.HindsightExperienceReplayWrapper(replay_buffer,
n_sampled_goal,
goal_selection_strategy,
wrapped_env)
Wrapper around a replay buffer in order to use HER. This implementation is inspired by to the one found in
https://github.com/NervanaSystems/coach/.
Parameters
• replay_buffer – (ReplayBuffer)
• n_sampled_goal – (int) The number of artificial transitions to generate for each actual
transition
• goal_selection_strategy – (GoalSelectionStrategy) The method that will be used
to generate the goals for the artificial transitions.
• wrapped_env – (HERGoalEnvWrapper) the GoalEnv wrapped using HERGoalEn-
vWrapper, that enables to convert observation to dict, and vice versa
add(obs_t, action, reward, obs_tp1, done, info)
add a new transition to the buffer
Parameters
• info – (dict) extra values used to compute reward
can_sample(n_samples)
Check if n_samples samples can be sampled from the buffer.
Parameters n_samples – (int)
Returns (bool)

1.26 PPO1
The Proximal Policy Optimization algorithm combines ideas from A2C (having multiple workers) and TRPO (it uses
a trust region to improve the actor).
The main idea is that after an update, the new policy should be not too far from the old policy. For that, ppo uses
clipping to avoid too large update.
Note: PPO1 requires OpenMPI. If OpenMPI isn’t enabled, then PPO1 isn’t imported into the stable_baselines
module.
Note: PPO1 uses MPI for multiprocessing unlike PPO2, which uses vectorized environments. PPO2 is the imple-
mentation OpenAI made for GPU.
1.26.1 Notes

• Clear explanation of PPO on Arxiv Insights channel: https://www.youtube.com/watch?v=5P7I-xPq8u8
• OpenAI blog post: https://blog.openai.com/openai-baselines-ppo/
• mpirun -np 8 python -m stable_baselines.ppo1.run_atari runs the algorithm for 40M
frames = 10M timesteps on an Atari game. See help (-h) for more options.
• python -m stable_baselines.ppo1.run_mujoco runs the algorithm for 1M frames on a Mujoco
environment.
• Train mujoco 3d humanoid (with optimal-ish hyperparameters): mpirun -np 16 python -m
stable_baselines.ppo1.run_humanoid --model-path=/path/to/model
• Render the 3d humanoid: python -m stable_baselines.ppo1.run_humanoid --play
--model-path=/path/to/model
1.26.2 Can I use?
• Gym spaces:

Discrete X X
Box X X
MultiDiscrete X X
MultiBinary X X
1.26. PPO1 129

1.26.3 Example
import gym


model.save("ppo1_cartpole")
model = PPO1.load("ppo1_cartpole")
obs = env.reset()
while True:
env.render()
1.26.4 Parameters
class stable_baselines.ppo1.PPO1(policy, env, gamma=0.99, timesteps_per_actorbatch=256,

clip_param=0.2, entcoeff=0.01, optim_epochs=4, op-
tim_stepsize=0.001, optim_batchsize=64, lam=0.95,
adam_epsilon=1e-05, schedule=’linear’, verbose=0,
tensorboard_log=None, _init_setup_model=True, pol-
n_cpu_tf_sess=1)
Proximal Policy Optimization algorithm (MPI version). Paper: https://arxiv.org/abs/1707.06347
Parameters
be str)
stmPolicy, . . . )
• timesteps_per_actorbatch – (int) timesteps per actor per update
• clip_param – (float) clipping parameter epsilon
• entcoeff – (float) the entropy loss weight
• optim_epochs – (float) the optimizer’s number of epochs
• optim_stepsize – (float) the optimizer’s stepsize
• optim_batchsize – (int) the optimizer’s the batch size
• lam – (float) advantage estimation

• schedule – (str) The type of scheduler for the learning rate update (‘linear’, ‘constant’,
‘double_linear_con’, ‘middle_drop’ or ‘double_middle_drop’)
instance
explanation
Parameters
distribution)
get_env()
1.26. PPO1 131

get_parameters()
learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’PPO1’, re-
Parameters
more details.
(used in logging)
Parameters
Parameters


Defaults to True.

Parameters
cies)
Parameters
Parameters
set_env(env)
change the seeds.
setup_model()
1.26. PPO1 133

From timestep 0
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• new_tb_log
• writer
• policy
• env
• horizon
• reward_giver
• gail
• step
• cur_ep_ret
• current_it_len
• current_ep_len
• cur_ep_true_ret
• ep_true_rets
• ep_rets
• ep_lens
• observations
• true_rewards
• rewards
• vpreds
• episode_starts
• dones
• actions
• states
• episode_start
• done
• vpred
• _
• i
• clipped_action
• reward
• true_reward
• info
• action
• observation
After the first episode termination

• maybe_ep_info

1.27 PPO2
The Proximal Policy Optimization algorithm combines ideas from A2C (having multiple workers) and TRPO (it uses
a trust region to improve the actor).
The main idea is that after an update, the new policy should be not too far from the old policy. For that, PPO uses
clipping to avoid too large update.
Note: PPO2 is the implementation of OpenAI made for GPU. For multiprocessing, it uses vectorized environments
compared to PPO1 which uses MPI.
Note: PPO2 contains several modifications from the original algorithm not documented by OpenAI: value function
is also clipped and advantages are normalized.
1.27.1 Notes

• Clear explanation of PPO on Arxiv Insights channel: https://www.youtube.com/watch?v=5P7I-xPq8u8
• python -m stable_baselines.ppo2.run_atari runs the algorithm for 40M frames = 10M
• python -m stable_baselines.ppo2.run_mujoco runs the algorithm for 1M frames on a Mu-
joco environment.
1.27.2 Can I use?
• Gym spaces:

Discrete X X
Box X X
MultiDiscrete X X
MultiBinary X X
1.27.3 Example
Train a PPO agent on CartPole-v1 using 4 processes.
import gym

1.27. PPO2 135



model.save("ppo2_cartpole")
model = PPO2.load("ppo2_cartpole")
# Enjoy trained agent

obs = env.reset()
while True:
env.render()
1.27.4 Parameters
class stable_baselines.ppo2.PPO2(policy, env, gamma=0.99, n_steps=128, ent_coef=0.01,

learning_rate=0.00025, vf_coef=0.5, max_grad_norm=0.5,
lam=0.95, nminibatches=4, noptepochs=4,
cliprange=0.2, cliprange_vf=None, verbose=0, ten-
sorboard_log=None, _init_setup_model=True, pol-
n_cpu_tf_sess=None)
Proximal Policy Optimization algorithm (GPU version). Paper: https://arxiv.org/abs/1707.06347
Parameters
stmPolicy, . . . )
be str)
• ent_coef – (float) Entropy coefficient for the loss calculation
• learning_rate – (float or callable) The learning rate, it can be a function
• vf_coef – (float) Value function coefficient for the loss calculation
• max_grad_norm – (float) The maximum value for the gradient clipping
• lam – (float) Factor for trade-off of bias vs variance for Generalized Advantage Estimator
• nminibatches – (int) Number of training minibatches per update. For recurrent policies,
the number of environments run in parallel should be a multiple of nminibatches.
• noptepochs – (int) Number of epoch when optimizing the surrogate
• cliprange – (float or callable) Clipping parameter, it can be a function

• cliprange_vf – (float or callable) Clipping parameter for the value function, it can be
a function. This is a parameter specific to the OpenAI implementation. If None is passed
(default), then cliprange (that is used for the policy) will be used. IMPORTANT: this clip-
ping depends on the reward scaling. To deactivate value function clipping (and recover the
original PPO implementation), you have to pass a negative value (e.g. -1).
instance
explanation
Parameters
distribution)
get_env()
1.27. PPO2 137

get_parameters()
learn(total_timesteps, callback=None, log_interval=1, tb_log_name=’PPO2’, re-
Parameters
more details.
(used in logging)
Parameters

Parameters
Defaults to True.

Parameters
cies)
Parameters
Parameters
set_env(env)
1.27. PPO2 139

change the seeds.
setup_model()
From timestep 1
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• cliprange_vf
• new_tb_log
• writer
• t_first_start
• n_updates
• mb_obs
• mb_rewards
• mb_actions
• mb_values
• mb_dones
• mb_neglogpacs
• mb_states
• ep_infos
• actions
• values
• neglogpacs
• clipped_actions
• rewards
• infos
From timestep 1
• info
• maybe_ep_info
1.28 SAC
Soft Actor Critic (SAC) Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor.
SAC is the successor of Soft Q-Learning SQL and incorporates the double Q-learning trick from TD3. A key feature
of SAC, and a major difference with common RL algorithms, is that it is trained to maximize a trade-off between
expected return and entropy, a measure of randomness in the policy.

Warning: The SAC model does not support stable_baselines.common.policies because it uses
double q-values and value estimation, as a result it must use its own policy models (see SAC Policies).
Available Policies

(2 layers of 64)
(the nature CNN)
1.28.1 Notes

• OpenAI Spinning Guide for SAC: https://spinningup.openai.com/en/latest/algorithms/sac.html
• Original Implementation: https://github.com/haarnoja/sac
• Blog post on using SAC with real robots: https://bair.berkeley.edu/blog/2018/12/14/sac/
Note: In our implementation, we use an entropy coefficient (as in OpenAI Spinning or Facebook Horizon), which is
the equivalent to the inverse of reward scale in the original SAC paper. The main reason is that it avoids having too
high errors when updating the Q functions.
Note: The default policies for SAC differ a bit from others MlpPolicy: it uses ReLU instead of tanh activation, to
match the original paper
1.28.2 Can I use?
• Gym spaces:

Discrete X
Box X X
MultiDiscrete X
MultiBinary X
1.28. SAC 141

1.28.3 Example
import gym
import numpy as np
from stable_baselines.sac.policies import MlpPolicy

model = SAC(MlpPolicy, env, verbose=1)

model.learn(total_timesteps=50000, log_interval=10)
model.save("sac_pendulum")
model = SAC.load("sac_pendulum")
obs = env.reset()
while True:
env.render()
1.28.4 Parameters
class stable_baselines.sac.SAC(policy, env, gamma=0.99, learning_rate=0.0003,

buffer_size=50000, learning_starts=100, train_freq=1,
batch_size=64, tau=0.005, ent_coef=’auto’, tar-
get_update_interval=1, gradient_steps=1, tar-
get_entropy=’auto’, action_noise=None, ran-
dom_exploration=0.0, verbose=0, tensorboard_log=None,
full_tensorboard_log=False, seed=None, n_cpu_tf_sess=None)
Soft Actor-Critic (SAC) Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor,
This implementation borrows code from original implementation (https://github.com/haarnoja/sac) from Ope-
nAI Spinning Up (https://github.com/openai/spinningup) and from the Softlearning repo (https://github.com/
rail-berkeley/softlearning/) Paper: https://arxiv.org/abs/1801.01290 Introduction to SAC: https://spinningup.
openai.com/en/latest/algorithms/sac.html
Parameters
• policy – (SACPolicy or str) The policy model to use (MlpPolicy, CnnPolicy, LnMlpPol-
icy, . . . )
be str)
• learning_rate – (float or callable) learning rate for adam optimizer, the same learning
rate will be used for all networks (Q-Values, Actor and Value function) it can be a function
of the current progress (from 1 to 0)
• batch_size – (int) Minibatch size for each gradient update

• tau – (float) the soft update coefficient (“polyak update”, between 0 and 1)
• ent_coef – (str or float) Entropy regularization coefficient. (Equivalent to inverse of
reward scale in the original SAC paper.) Controlling exploration/exploitation trade-off. Set
it to ‘auto’ to learn it automatically (and ‘auto_0.1’ for using 0.1 as initial value)
• train_freq – (int) Update the model every train_freq steps.
learning starts
• target_update_interval – (int) update the target network every tar-
get_network_update_freq steps.
• gradient_steps – (int) How many gradient update after each step
• target_entropy – (str or float) target entropy when learning ent_coef (ent_coef =
‘auto’)
• action_noise – (ActionNoise) the action noise type (None by default), this can help for
hard exploration problem. Cf DDPG for the different action noise type.
greedy strategy) This is not needed for SAC normally but can help exploring when using
HER + SAC. This hack was present in the original OpenAI Baselines repo (DDPG + HER)
instance
Note: this has no effect on SAC logging for now
explanation
Parameters
1.28. SAC 143

distribution)
get_env()
get_parameters()
learn(total_timesteps, callback=None, log_interval=4, tb_log_name=’SAC’, re-
Parameters
more details.
(used in logging)
Parameters


Parameters
Defaults to True.

Parameters
cies)
Parameters
1.28. SAC 145

Parameters
Parameters
set_env(env)
change the seeds.
setup_model()
1.28.5 SAC Policies
class stable_baselines.sac.MlpPolicy(sess, ob_space, ac_space, n_env=1, n_steps=1,

n_batch=None, reuse=False, **_kwargs)
Parameters

action_ph
initial_state
is_discrete
Creates an actor object
Parameters
holder)
make_critics(obs=None, action=None, reuse=False, scope=’values_fn’, create_vf=True, cre-
ate_qf=True)
Creates the two Q-Values approximator along with the Value function
Parameters
holder)
• action – (TensorFlow Tensor) The action placeholder
• scope – (str) the scope name
• create_vf – (bool) Whether to create Value fn or not
• create_qf – (bool) Whether to create Q-Values fn or not
Returns ([tf.Tensor]) Mean, action and log probability
obs_ph
Returns the action probability params (mean, std) for a single step
Parameters
Returns ([float], [float])
processed_obs
1.28. SAC 147


Parameters
class stable_baselines.sac.LnMlpPolicy(sess, ob_space, ac_space, n_env=1, n_steps=1,
Parameters
action_ph
initial_state
is_discrete
Parameters
holder)
ate_qf=True)
Parameters

holder)
obs_ph
Parameters
processed_obs
Parameters
class stable_baselines.sac.CnnPolicy(sess, ob_space, ac_space, n_env=1, n_steps=1,
Parameters
1.28. SAC 149


action_ph
initial_state
is_discrete
Parameters
holder)
ate_qf=True)
Parameters
holder)
obs_ph
Parameters

processed_obs
Parameters
class stable_baselines.sac.LnCnnPolicy(sess, ob_space, ac_space, n_env=1, n_steps=1,
Parameters
action_ph
initial_state
is_discrete
Parameters
holder)
1.28. SAC 151


ate_qf=True)
Parameters
holder)
obs_ph
Parameters
processed_obs
Parameters
network:

import gym
from stable_baselines.sac.policies import FeedForwardPolicy


class CustomSACPolicy(FeedForwardPolicy):
super(CustomSACPolicy, self).__init__(*args, **kwargs,
layers=[128, 128, 128],
layer_norm=False,

model = SAC(CustomSACPolicy, env, verbose=1)

# Train the agent
1.28. SAC 153

From timestep 1
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• replay_wrapper
• new_tb_log
• writer
• current_lr
• start_time
• episode_rewards
• obs
• n_updates
• infos_values
• step
• unscaled_action
• action
• new_obs
• reward
• done
• info
From timestep 2
• obs_
• new_obs_
• reward_
• maybe_ep_info
• mean_reward
• num_episodes
After timestep train_freq steps

• mb_infos_vals
• grad_step
After timestep train_freq steps After at least

• frac
batch_size and learning_starts steps
1.29 TD3
Twin Delayed DDPG (TD3) Addressing Function Approximation Error in Actor-Critic Methods.
TD3 is a direct successor of DDPG and improves it using three major tricks: clipped double Q-Learning, delayed
policy update and target policy smoothing. We recommend reading OpenAI Spinning guide on TD3 to learn more
about those.

Warning: The TD3 model does not support stable_baselines.common.policies because it uses
double q-values estimation, as a result it must use its own policy models (see TD3 Policies).
Available Policies

(2 layers of 64)
(the nature CNN)
1.29.1 Notes
• Original paper: https://arxiv.org/pdf/1802.09477.pdf

• OpenAI Spinning Guide for TD3: https://spinningup.openai.com/en/latest/algorithms/td3.html
• Original Implementation: https://github.com/sfujim/TD3
Note: The default policies for TD3 differ a bit from others MlpPolicy: it uses ReLU instead of tanh activation, to
match the original paper
1.29.2 Can I use?
• Gym spaces:

Discrete X
Box X X
MultiDiscrete X
MultiBinary X
1.29.3 Example
import gym
import numpy as np
from stable_baselines import TD3

from stable_baselines.td3.policies import MlpPolicy
from stable_baselines.ddpg.noise import NormalActionNoise,
˓→OrnsteinUhlenbeckActionNoise
1.29. TD3 155

# The noise objects for TD3

action_noise = NormalActionNoise(mean=np.zeros(n_actions), sigma=0.1 * np.ones(n_
˓→actions))
model = TD3(MlpPolicy, env, action_noise=action_noise, verbose=1)

model.learn(total_timesteps=50000, log_interval=10)
model.save("td3_pendulum")
model = TD3.load("td3_pendulum")
obs = env.reset()
while True:
env.render()
1.29.4 Parameters
class stable_baselines.td3.TD3(policy, env, gamma=0.99, learning_rate=0.0003,

buffer_size=50000, learning_starts=100, train_freq=100,
gradient_steps=100, batch_size=128, tau=0.005, pol-
icy_delay=2, action_noise=None, target_policy_noise=0.2,
target_noise_clip=0.5, random_exploration=0.0, ver-
bose=0, tensorboard_log=None, _init_setup_model=True,
policy_kwargs=None, full_tensorboard_log=False, seed=None,
n_cpu_tf_sess=None)
Twin Delayed DDPG (TD3) Addressing Function Approximation Error in Actor-Critic Methods.
Original implementation: https://github.com/sfujim/TD3 Paper: https://arxiv.org/pdf/1802.09477.pdf Introduc-
tion to TD3: https://spinningup.openai.com/en/latest/algorithms/td3.html
Parameters
• policy – (TD3Policy or str) The policy model to use (MlpPolicy, CnnPolicy, LnMlpPol-
icy, . . . )
be str)
• learning_rate – (float or callable) learning rate for adam optimizer, the same learning
rate will be used for all networks (Q-Values and Actor networks) it can be a function of the
current progress (from 1 to 0)
• batch_size – (int) Minibatch size for each gradient update
• tau – (float) the soft update coefficient (“polyak update” of the target networks, between 0
and 1)

• policy_delay – (int) Policy and target networks will only be updated once every pol-
icy_delay steps per training steps. The Q values will be updated policy_delay more often
(update every training step).
• action_noise – (ActionNoise) the action noise type. Cf DDPG for the different action
noise type.
• target_policy_noise – (float) Standard deviation of Gaussian noise added to target
policy (smoothing noise)
• target_noise_clip – (float) Limit for absolute value of target policy smoothing noise.
• train_freq – (int) Update the model every train_freq steps.
learning starts
• gradient_steps – (int) How many gradient update after each step
greedy strategy) This is not needed for TD3 normally but can help exploring when using
HER + TD3. This hack was present in the original OpenAI Baselines repo (DDPG + HER)
instance
Note: this has no effect on TD3 logging for now
explanation
Parameters
1.29. TD3 157

distribution)
get_env()
get_parameters()
learn(total_timesteps, callback=None, log_interval=4, tb_log_name=’TD3’, re-
Parameters
more details.
(used in logging)
Parameters

Parameters
Defaults to True.

Parameters
cies)
Parameters
1.29. TD3 159

Parameters
Parameters
set_env(env)
change the seeds.
setup_model()
1.29.5 TD3 Policies
class stable_baselines.td3.MlpPolicy(sess, ob_space, ac_space, n_env=1, n_steps=1,

Parameters

action_ph
initial_state
is_discrete
Parameters
holder)
make_critics(obs=None, action=None, reuse=False, scope=’values_fn’)
Creates the two Q-Values approximator
Parameters
holder)
obs_ph
Parameters
processed_obs
Parameters
1.29. TD3 161


class stable_baselines.td3.LnMlpPolicy(sess, ob_space, ac_space, n_env=1, n_steps=1,
Parameters
action_ph
initial_state
is_discrete
Parameters
holder)
Parameters
holder)

obs_ph
Parameters
processed_obs
Parameters
class stable_baselines.td3.CnnPolicy(sess, ob_space, ac_space, n_env=1, n_steps=1,
Parameters
action_ph
initial_state
is_discrete
1.29. TD3 163


Parameters
holder)
Parameters
holder)
obs_ph
Parameters
processed_obs
Parameters
class stable_baselines.td3.LnCnnPolicy(sess, ob_space, ac_space, n_env=1, n_steps=1,

Parameters
action_ph
initial_state
is_discrete
Parameters
holder)
Parameters
holder)
obs_ph
Parameters
1.29. TD3 165


processed_obs
Parameters
network:
import gym
import numpy as np
from stable_baselines import TD3

from stable_baselines.td3.policies import FeedForwardPolicy
from stable_baselines.ddpg.noise import NormalActionNoise,
˓→OrnsteinUhlenbeckActionNoise
# Custom MLP policy with two layers

class CustomTD3Policy(FeedForwardPolicy):
super(CustomTD3Policy, self).__init__(*args, **kwargs,
layers=[400, 300],
layer_norm=False,

# The noise objects for TD3

action_noise = NormalActionNoise(mean=np.zeros(n_actions), sigma=0.1 * np.ones(n_
˓→actions))
model = TD3(CustomTD3Policy, env, action_noise=action_noise, verbose=1)

# Train the agent

From timestep 1
• self
• total_timesteps
• callback
• log_interval
• tb_log_name
• replay_wrapper
• new_tb_log
• writer
• current_lr
• start_time
• episode_rewards
• obs
• n_updates
• infos_values
• step
• unscaled_action
• action
• new_obs
• reward
• done
• info
From timestep 2
• obs_
• new_obs_
• reward_
• maybe_ep_info
• mean_reward
• num_episodes
After timestep train_freq steps

• mb_infos_vals
• grad_step
After timestep train_freq steps After at least batch_size

• frac
and learning_starts steps
1.29. TD3 167

1.30 TRPO
Trust Region Policy Optimization (TRPO) is an iterative approach for optimizing policies with guaranteed monotonic
improvement.
Note: TRPO requires OpenMPI. If OpenMPI isn’t enabled, then TRPO isn’t imported into the
stable_baselines module.
1.30.1 Notes

• mpirun -np 16 python -m stable_baselines.trpo_mpi.run_atari runs the algorithm for
40M frames = 10M timesteps on an Atari game. See help (-h) for more options.
• python -m stable_baselines.trpo_mpi.run_mujoco runs the algorithm for 1M timesteps on a
Mujoco environment.
1.30.2 Can I use?
• Gym spaces:

Discrete X X
Box X X
MultiDiscrete X X
MultiBinary X X
1.30.3 Example
import gym

from stable_baselines import TRPO
model = TRPO(MlpPolicy, env, verbose=1)

model.save("trpo_cartpole")
model = TRPO.load("trpo_cartpole")


obs = env.reset()
while True:
env.render()
1.30.4 Parameters
class stable_baselines.trpo_mpi.TRPO(policy, env, gamma=0.99, timesteps_per_batch=1024,

max_kl=0.01, cg_iters=10, lam=0.98, entco-
eff=0.0, cg_damping=0.01, vf_stepsize=0.0003,
vf_iters=3, verbose=0, tensorboard_log=None,
full_tensorboard_log=False, seed=None,
n_cpu_tf_sess=1)
Trust Region Policy Optimization (https://arxiv.org/abs/1502.05477)
Parameters
stmPolicy, . . . )
be str)
• gamma – (float) the discount value
• timesteps_per_batch – (int) the number of timesteps to run per batch (horizon)
• max_kl – (float) the Kullback-Leibler loss threshold
• cg_iters – (int) the number of iterations for the conjugate gradient calculation
• lam – (float) GAE factor
• entcoeff – (float) the weight for the entropy loss
• cg_damping – (float) the compute gradient dampening factor
• vf_stepsize – (float) the value function stepsize
• vf_iters – (int) the value function’s number iterations for learning
instance
1.30. TRPO 169


explanation
Parameters
distribution)
get_env()
get_parameters()
learn(total_timesteps, callback=None, log_interval=100, tb_log_name=’TRPO’, re-
Parameters

more details.
(used in logging)
Parameters
Parameters
Defaults to True.

Parameters
1.30. TRPO 171

cies)
Parameters
Parameters
set_env(env)
change the seeds.
setup_model()

From timestep 0
• total_timesteps
• callback
• log_interval
• tb_log_name
• new_tb_log
• writer
• self
• policy
• env
• horizon
• reward_giver
• gail
• step
• cur_ep_ret
• current_it_len
• current_ep_len
• cur_ep_true_ret
• ep_true_rets
• ep_rets
• ep_lens
• observations
• true_rewards
• rewards
• vpreds
• episode_starts
• dones
• actions
• states
• episode_start
• done
• vpred
• clipped_action
• reward
• true_reward
• info
• action
• observation
• maybe_ep_info
1.31 Probability Distributions
Probability distributions used for the different action spaces:

• CategoricalProbabilityDistribution -> Discrete
• DiagGaussianProbabilityDistribution -> Box (continuous actions)
• MultiCategoricalProbabilityDistribution -> MultiDiscrete
• BernoulliProbabilityDistribution -> MultiBinary
1.31. Probability Distributions 173

The policy networks output parameters for the distributions (named flat in the methods). Actions are then sampled
from those distributions.
For instance, in the case of discrete actions. The policy network outputs probability of taking each action. The
CategoricalProbabilityDistribution allows to sample from it, computes the entropy, the negative log
probability (neglogp) and backpropagate the gradient.
In the case of continuous actions, a Gaussian distribution is used. The policy network outputs mean and (log) std of
the distribution (assumed to be a DiagGaussianProbabilityDistribution).
class stable_baselines.common.distributions.BernoulliProbabilityDistribution(logits)
entropy()
Returns Shannon’s entropy of the probability
Returns (float) the entropy
flatparam()
Return the direct probabilities
Returns ([float]) the probabilities
classmethod fromflat(flat)
Create an instance of this from new Bernoulli input
Parameters flat – ([float]) the Bernoulli input data
Returns (ProbabilityDistribution) the instance from the given Bernoulli input data
kl(other)
Calculates the Kullback-Leibler divergence from the given probability distribution
Parameters other – ([float]) the distribution to compare with
Returns (float) the KL divergence of the two distributions
mode()
Returns the probability
Returns (Tensorflow Tensor) the deterministic action
neglogp(x)
returns the of the negative log likelihood
Parameters x – (str) the labels of each index
Returns ([float]) The negative log likelihood of the distribution
sample()
returns a sample from the probability distribution
Returns (Tensorflow Tensor) the stochastic action
class stable_baselines.common.distributions.BernoulliProbabilityDistributionType(size)
param_shape()
returns the shape of the input parameters
Returns ([int]) the shape
proba_distribution_from_latent(pi_latent_vector, vf_latent_vector, init_scale=1.0,
init_bias=0.0)
returns the probability distribution from latent values
Parameters

• pi_latent_vector – ([float]) the latent pi values

• vf_latent_vector – ([float]) the latent vf values
• init_scale – (float) the initial scale of the distribution
• init_bias – (float) the initial bias of the distribution
Returns (ProbabilityDistribution) the instance of the ProbabilityDistribution associated
probability_distribution_class()
returns the ProbabilityDistribution class of this type
Returns (Type ProbabilityDistribution) the probability distribution class associated
sample_dtype()
returns the type of the sampling
Returns (type) the type
sample_shape()
returns the shape of the sampling
class stable_baselines.common.distributions.CategoricalProbabilityDistribution(logits)
entropy()
flatparam()
Create an instance of this from new logits values
Parameters flat – ([float]) the categorical logits input
Returns (ProbabilityDistribution) the instance from the given categorical input
kl(other)
mode()
neglogp(x)
sample()

class stable_baselines.common.distributions.CategoricalProbabilityDistributionType(n_cat)
param_shape()
init_bias=0.0)
Parameters
sample_dtype()
sample_shape()
class stable_baselines.common.distributions.DiagGaussianProbabilityDistribution(flat)
entropy()
flatparam()
Create an instance of this from new multivariate Gaussian input
Parameters flat – ([float]) the multivariate Gaussian input data
Returns (ProbabilityDistribution) the instance from the given multivariate Gaussian input data
kl(other)
mode()


neglogp(x)
sample()
class stable_baselines.common.distributions.DiagGaussianProbabilityDistributionType(size)
param_shape()
proba_distribution_from_flat(flat)
returns the probability distribution from flat probabilities
Parameters flat – ([float]) the flat probabilities
init_bias=0.0)
Parameters
sample_dtype()
sample_shape()
class stable_baselines.common.distributions.MultiCategoricalProbabilityDistribution(nvec,
flat)
entropy()

flatparam()
Create an instance of this from new logits values
Parameters flat – ([float]) the multi categorical logits input
Returns (ProbabilityDistribution) the instance from the given multi categorical input
kl(other)
mode()
neglogp(x)
sample()
class stable_baselines.common.distributions.MultiCategoricalProbabilityDistributionType(n_v
param_shape()
Returns the probability distribution from flat probabilities flat: flattened vector of parameters of probability
distribution
init_bias=0.0)
Parameters

sample_dtype()
sample_shape()
class stable_baselines.common.distributions.ProbabilityDistribution
Base class for describing a probability distribution.
entropy()
flatparam()
kl(other)
logp(x)
returns the of the log likelihood
Returns ([float]) The log likelihood of the distribution
mode()
neglogp(x)
sample()
class stable_baselines.common.distributions.ProbabilityDistributionType
Parametrized family of probability distributions
param_placeholder(prepend_shape, name=None)
returns the TensorFlow placeholder for the input parameters
Parameters

• prepend_shape – ([int]) the prepend shape

• name – (str) the placeholder name
Returns (TensorFlow Tensor) the placeholder
param_shape()
Returns the probability distribution from flat probabilities flat: flattened vector of parameters of probability
distribution
init_bias=0.0)
Parameters
sample_dtype()
sample_placeholder(prepend_shape, name=None)
returns the TensorFlow placeholder for the sampling
Parameters
• prepend_shape – ([int]) the prepend shape
• name – (str) the placeholder name
Returns (TensorFlow Tensor) the placeholder
sample_shape()
stable_baselines.common.distributions.make_proba_dist_type(ac_space)
return an instance of ProbabilityDistributionType for the correct type of action space
Parameters ac_space – (Gym Space) the input action space
Returns (ProbabilityDistributionType) the appropriate instance of a ProbabilityDistributionType

stable_baselines.common.distributions.shape_el(tensor, index)
get the shape of a TensorFlow Tensor element
Parameters
• tensor – (TensorFlow Tensor) the input tensor
• index – (int) the element
1.32 Tensorflow Utils

stable_baselines.common.tf_util.avg_norm(tensor)
Return an average of the L2 normalization of the batch
Parameters tensor – (TensorFlow Tensor) The input tensor
Returns (TensorFlow Tensor) Average L2 normalization of the batch
stable_baselines.common.tf_util.batch_to_seq(tensor_batch, n_batch, n_steps, flat=False)
Transform a batch of Tensors, into a sequence of Tensors for recurrent policies
Parameters
• tensor_batch – (TensorFlow Tensor) The input tensor to unroll
• flat – (bool) If the input Tensor is flat
Returns (TensorFlow Tensor) sequence of Tensors for recurrent policies
stable_baselines.common.tf_util.calc_entropy(logits)
Calculates the entropy of the output values of the network
Parameters logits – (TensorFlow Tensor) The input probability for each action
Returns (TensorFlow Tensor) The Entropy of the output values of the network
stable_baselines.common.tf_util.check_shape(tensors, shapes)
Verifies the tensors match the given shape, will raise an error if the shapes do not match
Parameters
• tensors – ([TensorFlow Tensor]) The tensors that should be checked
• shapes – ([list]) The list of shapes for each tensor
stable_baselines.common.tf_util.flatgrad(loss, var_list, clip_norm=None)
calculates the gradient and flattens it
Parameters
• loss – (float) the loss value
• var_list – ([TensorFlow Tensor]) the variables
• clip_norm – (float) clip the gradients (disabled if None)
Returns ([TensorFlow Tensor]) flattened gradient
1.32. Tensorflow Utils 181

stable_baselines.common.tf_util.function(inputs, outputs, updates=None, givens=None)

Take a bunch of tensorflow placeholders and expressions computed based on those placeholders and produces
f(inputs) -> outputs. Function f takes values to be fed to the input’s placeholders and produces the values of the
expressions in outputs. Just like a Theano function.
Input values can be passed in the same order as inputs or can be provided as kwargs based on placeholder name
(passed to constructor or accessible via placeholder.op.name).
Example:
>>> x = tf.placeholder(tf.int32, (), name="x")

>>> y = tf.placeholder(tf.int32, (), name="y")
>>> z = 3 * x + 2 * y
>>> lin = function([x, y], z, givens={y: 0})
>>> with single_threaded_session():
>>> initialize()
>>> assert lin(2) == 6
>>> assert lin(x=3) == 9
>>> assert lin(2, 2) == 10
Parameters
• inputs – (TensorFlow Tensor or Object with make_feed_dict) list of input arguments
• outputs – (TensorFlow Tensor) list of outputs or a single output to be returned from
function. Returned value will also have the same shape.
• updates – ([tf.Operation] or tf.Operation) list of update functions or single update func-
tion that will be run whenever the function is called. The return is ignored.
• givens – (dict) the values known for the output
stable_baselines.common.tf_util.get_globals_vars(name)
returns the trainable variables
Parameters name – (str) the scope
Returns ([TensorFlow Variable])
stable_baselines.common.tf_util.get_trainable_vars(name)
returns the trainable variables
Parameters name – (str) the scope
Returns ([TensorFlow Variable])
stable_baselines.common.tf_util.gradient_add(grad_1, grad_2, param, verbose=0)
Sum two gradients
Parameters
• grad_1 – (TensorFlow Tensor) The first gradient
• grad_2 – (TensorFlow Tensor) The second gradient
• param – (TensorFlow parameters) The trainable parameters
• verbose – (int) verbosity level
Returns (TensorFlow Tensor) the sum of the gradients
stable_baselines.common.tf_util.huber_loss(tensor, delta=1.0)
Reference: https://en.wikipedia.org/wiki/Huber_loss

Parameters
• tensor – (TensorFlow Tensor) the input value
• delta – (float) Huber loss delta value
Returns (TensorFlow Tensor) Huber loss output
stable_baselines.common.tf_util.in_session(func)
Wraps a function so that it is in a TensorFlow Session
Parameters func – (function) the function to wrap
Returns (function)
stable_baselines.common.tf_util.initialize(sess=None)
Initialize all the uninitialized variables in the global scope.
Parameters sess – (TensorFlow Session)
stable_baselines.common.tf_util.intprod(tensor)
calculates the product of all the elements in a list
Parameters tensor – ([Number]) the list of elements
Returns (int) the product truncated
stable_baselines.common.tf_util.is_image(tensor)
Check if a tensor has the shape of a valid image for tensorboard logging. Valid image: RGB, RGBD, GrayScale
Parameters tensor – (np.ndarray or tf.placeholder)
Returns (bool)
stable_baselines.common.tf_util.make_session(num_cpu=None, make_default=False,
graph=None)
Returns a session that will use <num_cpu> CPU’s only
Parameters
• num_cpu – (int) number of CPUs to use for TensorFlow
• make_default – (bool) if this should return an InteractiveSession or a normal Session
• graph – (TensorFlow Graph) the graph of the session
Returns (TensorFlow session)
stable_baselines.common.tf_util.mse(pred, target)
Returns the Mean squared error between prediction and target
Parameters
• pred – (TensorFlow Tensor) The predicted value
• target – (TensorFlow Tensor) The target value
Returns (TensorFlow Tensor) The Mean squared error between prediction and target
stable_baselines.common.tf_util.numel(tensor)
get TensorFlow Tensor’s number of elements
Parameters tensor – (TensorFlow Tensor) the input tensor
Returns (int) the number of elements
stable_baselines.common.tf_util.outer_scope_getter(scope, new_scope=”)
remove a scope layer for the getter
1.32. Tensorflow Utils 183

Parameters
• scope – (str) the layer to remove
• new_scope – (str) optional replacement name
Returns (function (function, str, *args, **kwargs): Tensorflow Tensor)
stable_baselines.common.tf_util.q_explained_variance(q_pred, q_true)
Calculates the explained variance of the Q value
Parameters
• q_pred – (TensorFlow Tensor) The predicted Q value
• q_true – (TensorFlow Tensor) The expected Q value
Returns (TensorFlow Tensor) the explained variance of the Q value
stable_baselines.common.tf_util.sample(logits)
Creates a sampling Tensor for non deterministic policies when using categorical distribution. It uses the Gumbel-
max trick: http://amid.fish/humble-gumbel
Parameters logits – (TensorFlow Tensor) The input probability for each action
Returns (TensorFlow Tensor) The sampled action
stable_baselines.common.tf_util.seq_to_batch(tensor_sequence, flat=False)
Transform a sequence of Tensors, into a batch of Tensors for recurrent policies
Parameters
• tensor_sequence – (TensorFlow Tensor) The input tensor to batch
• flat – (bool) If the input Tensor is flat
Returns (TensorFlow Tensor) batch of Tensors for recurrent policies
stable_baselines.common.tf_util.single_threaded_session(make_default=False,
graph=None)
Returns a session which will only use a single CPU
Parameters
• make_default – (bool) if this should return an InteractiveSession or a normal Session
• graph – (TensorFlow Graph) the graph of the session
Returns (TensorFlow session)
stable_baselines.common.tf_util.total_episode_reward_logger(rew_acc, rewards,
masks, writer, steps)
calculates the cumulated episode reward, and prints to tensorflow log the output
Parameters
• rew_acc – (np.array float) the total running reward
• rewards – (np.array float) the rewards
• masks – (np.array bool) the end of episodes
• writer – (TensorFlow Session.writer) the writer to log to
• steps – (int) the current timestep
Returns (np.array float) the updated total running reward
Returns (np.array float) the updated total running reward

stable_baselines.common.tf_util.var_shape(tensor)
get TensorFlow Tensor shape
Parameters tensor – (TensorFlow Tensor) the input tensor
1.33 Command Utils
Helpers for scripts like run_atari.py.

stable_baselines.common.cmd_util.arg_parser()
Create an empty argparse.ArgumentParser.
Returns (ArgumentParser)
stable_baselines.common.cmd_util.atari_arg_parser()
Create an argparse.ArgumentParser for run_atari.py.
Returns (ArgumentParser) parser {’–env’: ‘BreakoutNoFrameskip-v4’, ‘–seed’: 0, ‘–num-
timesteps’: int(1e7)}
stable_baselines.common.cmd_util.make_atari_env(env_id, num_env, seed, wrap-
per_kwargs=None, start_index=0,
allow_early_resets=True,
start_method=None,
use_subprocess=False)
Create a wrapped, monitored VecEnv for Atari.
Parameters
• env_id – (str) the environment ID
• num_env – (int) the number of environment you wish to have in subprocesses
• seed – (int) the initial seed for RNG
• wrapper_kwargs – (dict) the parameters for wrap_deepmind function
• start_index – (int) start rank index
• allow_early_resets – (bool) allows early reset of the environment
• start_method – (str) method used to start the subprocesses. See SubprocVecEnv doc
for more information
• use_subprocess – (bool) Whether to use SubprocVecEnv or DummyVecEnv when
num_env > 1, DummyVecEnv is usually faster. Default: False
Returns (VecEnv) The atari environment
stable_baselines.common.cmd_util.make_mujoco_env(env_id, seed, al-
low_early_resets=True)
Create a wrapped, monitored gym.Env for MuJoCo.
Parameters
Returns (Gym Environment) The mujoco environment
1.33. Command Utils 185

stable_baselines.common.cmd_util.make_robotics_env(env_id, seed, rank=0, al-

low_early_resets=True)
Create a wrapped, monitored gym.Env for MuJoCo.
Parameters
• rank – (int) the rank of the environment (for logging)
Returns (Gym Environment) The robotic environment
stable_baselines.common.cmd_util.make_vec_env(env_id, n_envs=1, seed=None,
start_index=0, moni-
tor_dir=None, wrapper_class=None,
env_kwargs=None, vec_env_cls=None,
vec_env_kwargs=None)
Create a wrapped, monitored VecEnv. By default it uses a DummyVecEnv which is usually faster than a Sub-
procVecEnv.
Parameters
• env_id – (str or Type[gym.Env]) the environment ID or the environment class
• n_envs – (int) the number of environments you wish to have in parallel
• seed – (int) the initial seed for the random number generator
• start_index – (int) start rank index
• monitor_dir – (str) Path to a folder where the monitor files will be saved. If None, no
file will be written, however, the env will still be wrapped in a Monitor wrapper to provide
additional information about training.
• wrapper_class – (gym.Wrapper or callable) Additional wrapper to use on the environ-
ment. This can also be a function with single argument that wraps the environment in many
things.
• env_kwargs – (dict) Optional keyword argument to pass to the env constructor
• vec_env_cls – (Type[VecEnv]) A custom VecEnv class constructor. Default: None.
• vec_env_kwargs – (dict) Keyword arguments to pass to the VecEnv class constructor.
Returns (VecEnv) The wrapped environment
stable_baselines.common.cmd_util.mujoco_arg_parser()
Create an argparse.ArgumentParser for run_mujoco.py.
Returns (ArgumentParser) parser {’–env’: ‘Reacher-v2’, ‘–seed’: 0, ‘–num-timesteps’: int(1e6),
‘–play’: False}
stable_baselines.common.cmd_util.robotics_arg_parser()
Create an argparse.ArgumentParser for run_mujoco.py.
Returns (ArgumentParser) parser {’–env’: ‘FetchReach-v0’, ‘–seed’: 0, ‘–num-timesteps’:
int(1e6)}

1.34 Schedules
Schedules are used as hyperparameter for most of the algorithms, in order to change value of a parameter over time
(usually the learning rate).
This file is used for specifying various schedules that evolve over time throughout the execution of the algorithm, such
as:
• learning rate for the optimizer
• exploration epsilon for the epsilon greedy exploration strategy
• beta parameter for beta parameter in prioritized replay
Each schedule has a function value(t) which returns the current value of the parameter given the timestep t of the
optimization procedure.
class stable_baselines.common.schedules.ConstantSchedule(value)
Value remains constant over time.
Parameters value – (float) Constant value of the schedule
value(step)
Value of the schedule for a given timestep
Parameters step – (int) the timestep
Returns (float) the output value for the given timestep
class stable_baselines.common.schedules.LinearSchedule(schedule_timesteps, final_p,
initial_p=1.0)
Linear interpolation between initial_p and final_p over schedule_timesteps. After this many timesteps pass
final_p is returned.
Parameters
• schedule_timesteps – (int) Number of timesteps for which to linearly anneal initial_p
to final_p
• initial_p – (float) initial output value
• final_p – (float) final output value
value(step)
class stable_baselines.common.schedules.PiecewiseSchedule(endpoints, interpo-
lation=<function
linear_interpolation>,
outside_value=None)
Piecewise schedule.
Parameters
• endpoints – ([(int, int)]) list of pairs (time, value) meaning that schedule should output
value when t==time. All the values for time must be sorted in an increasing order. When t
is between two times, e.g. (time_a, value_a) and (time_b, value_b), such that time_a <= t <
time_b then value outputs interpolation(value_a, value_b, alpha) where alpha is a fraction
of time passed between time_a and time_b for time t.
1.34. Schedules 187

• interpolation – (lambda (float, float, float): float) a function that takes value to the left
and to the right of t according to the endpoints. Alpha is the fraction of distance from left
endpoint to right endpoint that t has covered. See linear_interpolation for example.
• outside_value – (float) if the value is requested outside of all the intervals specified in
endpoints this value is returned. If None then AssertionError is raised when outside value is
requested.
value(step)
stable_baselines.common.schedules.constant(_)
Returns a constant value for the Scheduler
Parameters _ – ignored
Returns (float) 1
stable_baselines.common.schedules.constfn(val)
Create a function that returns a constant It is useful for learning rate schedule (to avoid code duplication)
Parameters val – (float)
Returns (function)
stable_baselines.common.schedules.double_linear_con(progress)
Returns a linear value (x2) with a flattened tail for the Scheduler
Parameters progress – (float) Current progress status (in [0, 1])
Returns (float) 1 - progress*2 if (1 - progress*2) >= 0.125 else 0.125
stable_baselines.common.schedules.double_middle_drop(progress)
Returns a linear value with two drops near the middle to a constant value for the Scheduler
Returns (float) if 0.75 <= 1 - p: 1 - p, if 0.25 <= 1 - p < 0.75: 0.75, if 1 - p < 0.25: 0.125
stable_baselines.common.schedules.get_schedule_fn(value_schedule)
Transform (if needed) learning rate and clip range to callable.
Parameters value_schedule – (callable or float)
Returns (function)
stable_baselines.common.schedules.linear_interpolation(left, right, alpha)
Linear interpolation between left and right.
Parameters
• left – (float) left boundary
• right – (float) right boundary
• alpha – (float) coeff in [0, 1]
Returns (float)
stable_baselines.common.schedules.linear_schedule(progress)
Returns a linear value for the Scheduler

Returns (float) 1 - progress

stable_baselines.common.schedules.middle_drop(progress)
Returns a linear value with a drop near the middle to a constant value for the Scheduler
Returns (float) 1 - progress if (1 - progress) >= 0.75 else 0.075
1.35 Evaluation Helper
stable_baselines.common.evaluation.evaluate_policy(model: BaseRLModel, env:

Union[gym.core.Env, sta-
ble_baselines.common.vec_env.base_vec_env.VecEnv],
n_eval_episodes: int = 10,
deterministic: bool = True,
render: bool = False, call-
back: Optional[Callable]
= None, reward_threshold:
Optional[float] = None, re-
turn_episode_rewards: bool =
False) → Union[Tuple[float,
float], Tuple[List[float], List[int]]]
Runs policy for n_eval_episodes episodes and returns average reward. This is made to work only with
one env.
Parameters
• model – (BaseRLModel) The RL agent you want to evaluate.
• env – (gym.Env or VecEnv) The gym environment. In the case of a VecEnv this must
contain only one environment.
• n_eval_episodes – (int) Number of episode to evaluate the agent
• deterministic – (bool) Whether to use deterministic or stochastic actions
• render – (bool) Whether to render the environment or not
• callback – (callable) callback function to do additional checks, called after each step.
• reward_threshold – (float) Minimum expected reward per episode, this will raise an
error if the performance is not met
• return_episode_rewards – (Optional[float]) If True, a list of reward per episode will
be returned instead of the mean.
Returns (float, float) Mean reward per episode, std of reward per episode returns ([float], [int]) when
return_episode_rewards is True
1.36 Gym Environment Checker
stable_baselines.common.env_checker.check_env(env: gym.core.Env, warn: bool = True,

skip_render_check: bool = True) →
None
Check that an environment follows Gym API. This is particularly useful when using a custom environment.
Please take a look at https://github.com/openai/gym/blob/master/gym/core.py for more information about the
API.
1.35. Evaluation Helper 189

It also optionally check that the environment is compatible with Stable-Baselines.

Parameters
• env – (gym.Env) The Gym environment that will be checked
• warn – (bool) Whether to output additional warnings mainly related to the interaction with
Stable Baselines
• skip_render_check – (bool) Whether to skip the checks for the render method. True
by default (useful for the CI)
1.37 Monitor Wrapper
class stable_baselines.bench.monitor.Monitor(env: gym.core.Env, filename: Optional[str],

allow_early_resets: bool = True, re-
set_keywords=(), info_keywords=())
A monitor wrapper for Gym environments, it is used to know the episode reward, length, time and other data.
Parameters
• env – (gym.Env) The environment
• filename – (Optional[str]) the location to save a log file, can be None for no log
• allow_early_resets – (bool) allows the reset of the environment before it is done
• reset_keywords – (tuple) extra keywords for the reset call, if extra parameters are
needed at reset
• info_keywords – (tuple) extra information to log, from the information return of envi-
ronment.step
close()
Closes the environment
get_episode_lengths() → List[int]
Returns the number of timesteps of all the episodes
Returns ([int])
get_episode_rewards() → List[float]
Returns the rewards of all the episodes
Returns ([float])
get_episode_times() → List[float]
Returns the runtime in seconds of all the episodes
Returns ([float])
get_total_steps() → int
Returns the total number of timesteps
Returns (int)
reset(**kwargs) → numpy.ndarray
Calls the Gym environment reset. Can only be called if the environment is over, or if allow_early_resets
is True
Parameters kwargs – Extra keywords saved for the next episode. only if defined by re-
set_keywords

Returns (np.ndarray) the first observation of the environment

step(action: numpy.ndarray) → Tuple[numpy.ndarray, float, bool, Dict[Any, Any]]
Step the environment with the given action
Parameters action – (np.ndarray) the action
Returns (Tuple[np.ndarray, float, bool, Dict[Any, Any]]) observation, reward, done, information
stable_baselines.bench.monitor.get_monitor_files(path: str) → List[str]
get all the monitor files in the given path
Parameters path – (str) the logging folder
Returns ([str]) the log files
stable_baselines.bench.monitor.load_results(path: str) → pandas.core.frame.DataFrame
Load all Monitor logs from a given directory path matching *monitor.csv and *monitor.json
Parameters path – (str) the directory path containing the log file(s)
Returns (pandas.DataFrame) the logged data
1.38 Changelog
For download links, please look at Github release page.
1.38.1 Release 2.10.2 (2021-04-05)
Warning: This package is in maintenance mode, please use Stable-Baselines3 (SB3) for an up-to-date version.
You can find a migration guide in SB3 documentation.
Breaking Changes:
New Features:
• EvalCallback now works also for recurrent policies (@mily20001)
Bug Fixes:
• Fixed calculation of the log probability of Diagonal Gaussian distribution when using
action_probability() method (@SVJayanthi, @sunshineclt)
• Fixed docker image build (@anj1)
Deprecations:
Others:
• Faster tests, switched to GitHub CI
1.38. Changelog 191

Documentation:
• Added stable-baselines-tf2 link on Projects page. (@sophiagu)

• Fixed a typo in stable_baselines.common.env_checker.check_env (@OGordon100)
1.38.2 Release 2.10.1 (2020-08-05)
Bug fixes release
Breaking Changes:
• render() method of VecEnvs now only accept one argument: mode
New Features:
• Added momentum parameter to A2C for the embedded RMSPropOptimizer (@kantneel)

• ActionNoise is now an abstract base class and implements __call__, NormalActionNoise and
OrnsteinUhlenbeckActionNoise have return types (@PartiallyTyped)
• HER now passes info dictionary to compute_reward, allowing for the computation of rewards that are indepen-
dent of the goal (@tirafesi)
Bug Fixes:
• Fixed DDPG sampling empty replay buffer when combined with HER (@tirafesi)
• Fixed a bug in HindsightExperienceReplayWrapper, where the openai-gym signature for
compute_reward was not matched correctly (@johannes-dornheim)
• Fixed SAC/TD3 checking time to update on learn steps instead of total steps (@PartiallyTyped)
• Added **kwarg pass through for reset method in atari_wrappers.FrameStack (@PartiallyTyped)
• Fix consistency in setup_model() for SAC, target_entropy now uses self.action_space in-
stead of self.env.action_space (@PartiallyTyped)
• Fix reward threshold in test_identity.py
• Partially fix tensorboard indexing for PPO2 (@enderdead)
• Fixed potential bug in DummyVecEnv where copy() was used instead of deepcopy()
• Fixed a bug in GAIL where the dataloader was not available after saving, causing an error when using
CheckpointCallback
• Fixed a bug in SAC where any convolutional layers were not included in the target network parameters.
• Fixed render() method for VecEnvs
• Fixed seed()` method for SubprocVecEnv
• Fixed a bug callback.locals did not have the correct values (@PartiallyTyped)
• Fixed a bug in the close() method of SubprocVecEnv, causing wrappers further down in the wrapper
stack to not be closed. (@NeoExtended)

• Fixed a bug in the generate_expert_traj() method in record_expert.py when using a non-image

vectorized environment (@jbarsce)
• Fixed a bug in CloudPickleWrapper’s (used by VecEnvs) __setstate___ where loading was incorrectly
using pickle.loads (@shwang).
• Fixed a bug in SAC and TD3 where the log timesteps was not correct(@YangRui2015)
• Fixed a bug where the environment was reset twice when using evaluate_policy
Deprecations:
Others:
• Added version.txt to manage version number in an easier way

• Added .readthedocs.yml to install requirements with read the docs
• Added a test for seeding SubprocVecEnv` and rendering
Documentation:
• Fix typos (@caburu)

• Fix typos in PPO2 (@kvenkman)
• Removed stable_baselines\deepq\experiments\custom_cartpole.py (@aakash94)
• Added Google’s motion imitation project
• Added documentation page for monitor
• Fixed typos and update VecNormalize example to show normalization at test-time
• Fixed train_mountaincar description
• Added imitation baselines project
• Updated install instructions
• Added Slime Volleyball project (@hardmaru)
• Added a table of the variables accessible from the on_step function of the callbacks for each algorithm
(@PartiallyTyped)
• Fix typo in README.md (@ColinLeongUDRI)
• Fix typo in gail.rst (@roccivic)
1.38.3 Release 2.10.0 (2020-03-11)
Callback collection, cleanup and bug fixes
Breaking Changes:
• evaluate_policy now returns the standard deviation of the reward per episode as second return value
(instead of n_steps)
• evaluate_policy now returns as second return value a list of the episode lengths when
return_episode_rewards is set to True (instead of n_steps)
1.38. Changelog 193

• Callback are now called after each env.step() for consistency (it was called every n_steps before in
algorithm like A2C or PPO2)
• Removed unused code in common/a2c/utils.py (calc_entropy_softmax, make_path)
• Refactoring, including removed files and moving functions.
– Algorithms no longer import from each other, and common does not import from algorithms.
– a2c/utils.py removed and split into other files:
* common/tf_util.py: sample, calc_entropy, mse, avg_norm,

total_episode_reward_logger, q_explained_variance, gradient_add,
avg_norm, check_shape, seq_to_batch, batch_to_seq.
* common/tf_layers.py: conv, linear, lstm, _ln, lnlstm, conv_to_fc, ortho_init.

* a2c/a2c.py: discount_with_dones.
* acer/acer_simple.py: get_by_index, EpisodeStats.
* common/schedules.py: constant, linear_schedule, middle_drop,
double_linear_con, double_middle_drop, SCHEDULES, Scheduler.
– trpo_mpi/utils.py functions moved (traj_segment_generator moved to common/
runners.py, flatten_lists to common/misc_util.py).
– ppo2/ppo2.py functions moved (safe_mean to common/math_util.py, constfn and
get_schedule_fn to common/schedules.py).
– sac/policies.py function mlp moved to common/tf_layers.py.
– sac/sac.py function get_vars removed (replaced with tf.util.get_trainable_vars).
– deepq/replay_buffer.py renamed to common/buffers.py.
New Features:
• Parallelized updating and sampling from the replay buffer in DQN. (@flodorner)
• Docker build script, scripts/build_docker.sh, can push images automatically.
• Added callback collection
• Added unwrap_vec_normalize and sync_envs_normalization in the vec_env module to syn-
chronize two VecNormalize environment
• Added a seeding method for vectorized environments. (@NeoExtended)
• Added extend method to store batches of experience in ReplayBuffer. (@PartiallyTyped)
Bug Fixes:
• Fixed Docker images via scripts/build_docker.sh and Dockerfile: GPU image now contains
tensorflow-gpu, and both images have stable_baselines installed in developer mode at correct di-
rectory for mounting.
• Fixed Docker GPU run script, scripts/run_docker_gpu.sh, to work with new NVidia Container
Toolkit.
• Repeated calls to RLModel.learn() now preserve internal counters for some episode logging statistics that
used to be zeroed at the start of every call.

• Fix DummyVecEnv.render for num_envs > 1. This used to print a warning and then not render at all.
(@shwang)
• Fixed a bug in PPO2, ACER, A2C, and ACKTR where repeated calls to learn(total_timesteps) reset
the environment on every call, potentially biasing samples toward early episode timesteps. (@shwang)
• Fixed by adding lazy property ActorCriticRLModel.runner. Subclasses now use lazily-generated
self.runner instead of reinitializing a new Runner every time learn() is called.
• Fixed a bug in check_env where it would fail on high dimensional action spaces
• Fixed Monitor.close() that was not calling the parent method
• Fixed a bug in BaseRLModel when seeding vectorized environments. (@NeoExtended)
• Fixed num_timesteps computation to be consistent between algorithms (updated after env.step()) Only
TRPO and PPO1 update it differently (after synchronization) because they rely on MPI
• Fixed bug in TRPO with NaN standardized advantages (@richardwu)
• Fixed partial minibatch computation in ExpertDataset (@richardwu)
• Fixed normalization (with VecNormalize) for off-policy algorithms
• Fixed sync_envs_normalization to sync the reward normalization too
• Bump minimum Gym version (>=0.11)
Deprecations:
Others:
• Removed redundant return value from a2c.utils::total_episode_reward_logger. (@shwang)

• Cleanup and refactoring in common/identity_env.py (@shwang)
• Added a Makefile to simplify common development tasks (build the doc, type check, run the tests)
Documentation:
• Add dedicated page for callbacks

• Fixed example for creating a GIF (@KuKuXia)
• Change Colab links in the README to point to the notebooks repo
• Fix typo in Reinforcement Learning Tips and Tricks page. (@mmcenta)
1.38.4 Release 2.9.0 (2019-12-20)
Reproducible results, automatic ‘‘VecEnv‘‘ wrapping, env checker and more usability improvements
Breaking Changes:
• The seed argument has been moved from learn() method to model constructor in order to have reproducible
results
• allow_early_resets of the Monitor wrapper now default to True
1.38. Changelog 195

• make_atari_env now returns a DummyVecEnv by default (instead of a SubprocVecEnv) this usually

improves performance.
• Fix inconsistency of sample type, so that mode/sample function returns tensor of tf.int64 in CategoricalProba-
bilityDistribution/MultiCategoricalProbabilityDistribution (@seheevic)
New Features:
• Add n_cpu_tf_sess to model constructor to choose the number of threads used by Tensorflow
• Environments are automatically wrapped in a DummyVecEnv if needed when passing them to the model con-
structor
• Added stable_baselines.common.make_vec_env helper to simplify VecEnv creation
• Added stable_baselines.common.evaluation.evaluate_policy helper to simplify model
evaluation
• VecNormalize changes:
– Now supports being pickled and unpickled (@AdamGleave).
– New methods .normalize_obs(obs) and normalize_reward(rews) apply normalization to arbitrary
observation or rewards without updating statistics (@shwang)
– .get_original_reward() returns the unnormalized rewards from the most recent timestep
– .reset() now collects observation statistics (used to only apply normalization)
• Add parameter exploration_initial_eps to DQN. (@jdossgollin)
• Add type checking and PEP 561 compliance. Note: most functions are still not annotated, this will be a gradual
process.
• DDPG, TD3 and SAC accept non-symmetric action spaces. (@Antymon)
• Add check_env util to check if a custom environment follows the gym interface (@araffin and @justinkterry)
Bug Fixes:
• Fix seeding, so it is now possible to have deterministic results on cpu

• Fix a bug in DDPG where predict method with deterministic=False would fail
• Fix a bug in TRPO: mean_losses was not initialized causing the logger to crash when there was no gradients
(@MarvineGothic)
• Fix a bug in cmd_util from API change in recent Gym versions
• Fix a bug in DDPG, TD3 and SAC where warmup and random exploration actions would end up scaled in the
replay buffer (@Antymon)
Deprecations:
• nprocs (ACKTR) and num_procs (ACER) are deprecated in favor of n_cpu_tf_sess which is now
common to all algorithms
• VecNormalize: load_running_average and save_running_average are deprecated in favour of
using pickle.

Others:
• Add upper bound for Tensorflow version (<2.0.0).

• Refactored test to remove duplicated code
• Add pull request template
• Replaced redundant code in load_results (@jbulow)
• Minor PEP8 fixes in dqn.py (@justinkterry)
• Add a message to the assert in PPO2
• Update replay buffer doctring
• Fix VecEnv docstrings
Documentation:
• Add plotting to the Monitor example (@rusu24edward)

• Add Snake Game AI project (@pedrohbtp)
• Add note on the support Tensorflow versions.
• Remove unnecessary steps required for Windows installation.
• Remove DummyVecEnv creation when not needed
• Added make_vec_env to the examples to simplify VecEnv creation
• Add QuaRL project (@srivatsankrishnan)
• Add Pwnagotchi project (@evilsocket)
• Fix multiprocessing example (@rusu24edward)
• Fix result_plotter example
• Add JNRR19 tutorial (by @edbeeching, @hill-a and @araffin)
• Updated notebooks link
• Fix typo in algos.rst, “containes” to “contains” (@SyllogismRXS)
• Fix outdated source documentation for load_results
• Add PPO_CPP project (@Antymon)
• Add section on C++ portability of Tensorflow models (@Antymon)
• Update custom env documentation to reflect new gym API for the close() method (@justinkterry)
• Update custom env documentation to clarify what step and reset return (@justinkterry)
• Add RL tips and tricks for doing RL experiments
• Corrected lots of typos
• Add spell check to documentation if available
1.38.5 Release 2.8.0 (2019-09-29)
MPI dependency optional, new save format, ACKTR with continuous actions
1.38. Changelog 197

Breaking Changes:
• OpenMPI-dependent algorithms (PPO1, TRPO, GAIL, DDPG) are disabled in the default installation of sta-
ble_baselines. mpi4py is now installed as an extra. When mpi4py is not available, stable-baselines skips
imports of OpenMPI-dependent algorithms. See installation notes and Issue #430.
• SubprocVecEnv now defaults to a thread-safe start method, forkserver when available and otherwise
spawn. This may require application code be wrapped in if __name__ == '__main__'. You can re-
store previous behavior by explicitly setting start_method = 'fork'. See PR #428.
• Updated dependencies: tensorflow v1.8.0 is now required
• Removed checkpoint_path and checkpoint_freq argument from DQN that were not used
• Removed bench/benchmark.py that was not used
• Removed several functions from common/tf_util.py that were not used
• Removed ppo1/run_humanoid.py
New Features:
• important change Switch to using zip-archived JSON and Numpy savez for storing models for better support
across library/Python versions. (@Miffyli)
• ACKTR now supports continuous actions
• Add double_q argument to DQN constructor
Bug Fixes:
• Skip automatic imports of OpenMPI-dependent algorithms to avoid an issue where OpenMPI would cause
stable-baselines to hang on Ubuntu installs. See installation notes and Issue #430.
• Fix a bug when calling logger.configure() with MPI enabled (@keshaviyengar)
• set allow_pickle=True for numpy>=1.17.0 when loading expert dataset
• Fix a bug when using VecCheckNan with numpy ndarray as state. Issue #489. (@ruifeng96150)
Deprecations:
• Models saved with cloudpickle format (stable-baselines<=2.7.0) are now deprecated in favor of zip-archive
format for better support across Python/Tensorflow versions. (@Miffyli)
Others:
• Implementations of noise classes (AdaptiveParamNoiseSpec, NormalActionNoise,

OrnsteinUhlenbeckActionNoise) were moved from stable_baselines.ddpg.noise to
stable_baselines.common.noise. The API remains backward-compatible; for example from
stable_baselines.ddpg.noise import NormalActionNoise is still okay. (@shwang)
• Docker images were updated
• Cleaned up files in common/ folder and in acktr/ folder that were only used by old ACKTR version (e.g.
filter.py)
• Renamed acktr_disc.py to acktr.py

Documentation:
• Add WaveRL project (@jaberkow)

• Add Fenics-DRL project (@DonsetPG)
• Fix and rename custom policy names (@eavelardev)
• Add documentation on exporting models.
• Update maintainers list (Welcome to @Miffyli)
1.38.6 Release 2.7.0 (2019-07-31)
Twin Delayed DDPG (TD3) and GAE bug fix (TRPO, PPO1, GAIL)
Breaking Changes:
New Features:
• added Twin Delayed DDPG (TD3) algorithm, with HER support

• added support for continuous action spaces to action_probability, computing the PDF of a Gaussian
policy in addition to the existing support for categorical stochastic policies.
• added flag to action_probability to return log-probabilities.
• added support for python lists and numpy arrays in logger.writekvs. (@dwiel)
• the info dict returned by VecEnvs now include a terminal_observation key providing access to the last
observation in a trajectory. (@qxcv)
Bug Fixes:
• fixed a bug in traj_segment_generator where the episode_starts was wrongly recorded, resulting
in wrong calculation of Generalized Advantage Estimation (GAE), this affects TRPO, PPO1 and GAIL (thanks
to @miguelrass for spotting the bug)
• added missing property n_batch in BasePolicy.
Deprecations:
Others:
• renamed some keys in traj_segment_generator to be more meaningful

• retrieve unnormalized reward when using Monitor wrapper with TRPO, PPO1 and GAIL to display them in the
logs (mean episode reward)
• clean up DDPG code (renamed variables)
Documentation:
• doc fix for the hyperparameter tuning command in the rl zoo

• added an example on how to log additional variable with tensorboard and a callback
1.38. Changelog 199

1.38.7 Release 2.6.0 (2019-06-12)
Hindsight Experience Replay (HER) - Reloaded | get/load parameters
Breaking Changes:
• breaking change removed stable_baselines.ddpg.memory in favor of stable_baselines.

deepq.replay_buffer (see fix below)
Breaking Change: DDPG replay buffer was unified with DQN/SAC replay buffer. As a result, when loading a DDPG
model trained with stable_baselines<2.6.0, it throws an import error. You can fix that using:
import sys
import pkg_resources
import stable_baselines
# Fix for breaking change for DDPG buffer in v2.6.0

if pkg_resources.get_distribution("stable_baselines").version >= "2.6.0":
sys.modules['stable_baselines.ddpg.memory'] = stable_baselines.deepq.replay_buffer
stable_baselines.deepq.replay_buffer.Memory = stable_baselines.deepq.replay_
˓→buffer.ReplayBuffer
We recommend you to save again the model afterward, so the fix won’t be needed the next time the trained agent is
loaded.
New Features:
• revamped HER implementation: clean re-implementation from scratch, now supports DQN, SAC and DDPG
• add action_noise param for SAC, it helps exploration for problem with deceptive reward
• The parameter filter_size of the function conv in A2C utils now supports passing a list/tuple of two
integers (height and width), in order to have non-squared kernel matrix. (@yutingsz)
• add random_exploration parameter for DDPG and SAC, it may be useful when using HER + DDPG/SAC.
This hack was present in the original OpenAI Baselines DDPG + HER implementation.
• added load_parameters and get_parameters to base RL class. With these methods, users are able to
load and get parameters to/from existing model, without touching tensorflow. (@Miffyli)
• added specific hyperparameter for PPO2 to clip the value function (cliprange_vf)
• added VecCheckNan wrapper
Bug Fixes:
• bugfix for VecEnvWrapper.__getattr__ which enables access to class attributes inherited from parent
classes.
• fixed path splitting in TensorboardWriter._get_latest_run_id() on Windows machines
(@PatrickWalter214)
• fixed a bug where initial learning rate is logged instead of its placeholder in A2C.setup_model (@sc420)
• fixed a bug where number of timesteps is incorrectly updated and logged in A2C.learn and A2C.
_train_step (@sc420)

• fixed num_timesteps (total_timesteps) variable in PPO2 that was wrongly computed.

• fixed a bug in DDPG/DQN/SAC, when there were the number of samples in the replay buffer was lesser than
the batch size (thanks to @dwiel for spotting the bug)
• removed a2c.utils.find_trainable_params please use common.tf_util.
get_trainable_vars instead. find_trainable_params was returning all trainable variables,
discarding the scope argument. This bug was causing the model to save duplicated parameters (for DDPG and
SAC) but did not affect the performance.
Deprecations:
• deprecated memory_limit and memory_policy in DDPG, please use buffer_size instead. (will be
removed in v3.x.x)
Others:
• important change switched to using dictionaries rather than lists when storing parameters, with tensorflow
Variable names being the keys. (@Miffyli)
• removed unused dependencies (tdqm, dill, progressbar2, seaborn, glob2, click)
• removed get_available_gpus function which hadn’t been used anywhere (@Pastafarianist)
Documentation:
• added guide for managing NaN and inf

• updated ven_env doc
• misc doc updates
1.38.8 Release 2.5.1 (2019-05-04)
Bug fixes + improvements in the VecEnv

Warning: breaking changes when using custom policies
• doc update (fix example of result plotter + improve doc)
• fixed logger issues when stdout lacks read function
• fixed a bug in common.dataset.Dataset where shuffling was not disabled properly (it affects only PPO1
with recurrent policies)
• fixed output layer name for DDPG q function, used in pop-art normalization and l2 regularization of the critic
• added support for multi env recording to generate_expert_traj (@XMaster96)
• added support for LSTM model recording to generate_expert_traj (@XMaster96)
• GAIL: remove mandatory matplotlib dependency and refactor as subclass of TRPO (@kantneel and
@AdamGleave)
• added get_attr(), env_method() and set_attr() methods for all VecEnv. Those methods now all
accept indices keyword to select a subset of envs. set_attr now returns None rather than a list of
None. (@kantneel)
1.38. Changelog 201

• GAIL: gail.dataset.ExpertDataset supports loading from memory rather than file, and gail.
dataset.record_expert supports returning in-memory rather than saving to file.
• added support in VecEnvWrapper for accessing attributes of arbitrarily deeply nested instances of
VecEnvWrapper and VecEnv. This is allowed as long as the attribute belongs to exactly one of the nested
instances i.e. it must be unambiguous. (@kantneel)
• fixed bug where result plotter would crash on very short runs (@Pastafarianist)
• added option to not trim output of result plotter by number of timesteps (@Pastafarianist)
• clarified the public interface of BasePolicy and ActorCriticPolicy. Breaking change when using
custom policies: masks_ph is now called dones_ph, and most placeholders were made private: e.g. self.
value_fn is now self._value_fn
• support for custom stateful policies.
• fixed episode length recording in trpo_mpi.utils.traj_segment_generator (@GerardMaggi-
olino)
1.38.9 Release 2.5.0 (2019-03-28)
Working GAIL, pretrain RL models and hotfix for A2C with continuous actions
• fixed various bugs in GAIL
• added scripts to generate dataset for gail
• added tests for GAIL + data for Pendulum-v0
• removed unused utils file in DQN folder
• fixed a bug in A2C where actions were cast to int32 even in the continuous case
• added addional logging to A2C when Monitor wrapper is used
• changed logging for PPO2: do not display NaN when reward info is not present
• change default value of A2C lr schedule
• removed behavior cloning script
• added pretrain method to base class, in order to use behavior cloning on all models
• fixed close() method for DummyVecEnv.
• added support for Dict spaces in DummyVecEnv and SubprocVecEnv. (@AdamGleave)
• added support for arbitrary multiprocessing start methods and added a warning about SubprocVecEnv that are
not thread-safe by default. (@AdamGleave)
• added support for Discrete actions for GAIL
• fixed deprecation warning for tf: replaces tf.to_float() by tf.cast()
• fixed bug in saving and loading ddpg model when using normalization of obs or returns (@tperol)
• changed DDPG default buffer size from 100 to 50000.
• fixed a bug in ddpg.py in combined_stats for eval. Computed mean on eval_episode_rewards
and eval_qs (@keshaviyengar)
• fixed a bug in setup.py that would error on non-GPU systems without TensorFlow installed

1.38.10 Release 2.4.1 (2019-02-11)
Bug fixes and improvements

• fixed computation of training metrics in TRPO and PPO1
• added reset_num_timesteps keyword when calling train() to continue tensorboard learning curves
• reduced the size taken by tensorboard logs (added a full_tensorboard_log to enable full logging, which
was the previous behavior)
• fixed image detection for tensorboard logging
• fixed ACKTR for recurrent policies
• fixed gym breaking changes
• fixed custom policy examples in the doc for DQN and DDPG
• remove gym spaces patch for equality functions
• fixed tensorflow dependency: cpu version was installed overwritting tensorflow-gpu when present.
• fixed a bug in traj_segment_generator (used in ppo1 and trpo) where new was not updated. (spotted
by @junhyeokahn)
1.38.11 Release 2.4.0 (2019-01-17)
Soft Actor-Critic (SAC) and policy kwargs

• added Soft Actor-Critic (SAC) model
• fixed a bug in DQN where prioritized_replay_beta_iters param was not used
• fixed DDPG that did not save target network parameters
• fixed bug related to shape of true_reward (@abhiskk)
• fixed example code in documentation of tf_util:Function (@JohannesAck)
• added learning rate schedule for SAC
• fixed action probability for continuous actions with actor-critic models
• added optional parameter to action_probability for likelihood calculation of given action being taken.
• added more flexible custom LSTM policies
• added auto entropy coefficient optimization for SAC
• clip continuous actions at test time too for all algorithms (except SAC/DDPG where it is not needed)
• added a mean to pass kwargs to policy when creating a model (+ save those kwargs)
• fixed DQN examples in DQN folder
• added possibility to pass activation function for DDPG, DQN and SAC
1.38.12 Release 2.3.0 (2018-12-05)
• added support for storing model in file like object. (thanks to @ernestum)
• fixed wrong image detection when using tensorboard logging with DQN
• fixed bug in ppo2 when passing non callable lr after loading
1.38. Changelog 203

• fixed tensorboard logging in ppo2 when nminibatches=1

• added early stoppping via callback return value (@ernestum)
• added more flexible custom mlp policies (@ernestum)
1.38.13 Release 2.2.1 (2018-11-18)
• added VecVideoRecorder to record mp4 videos from environment.
1.38.14 Release 2.2.0 (2018-11-07)
• Hotfix for ppo2, the wrong placeholder was used for the value function
1.38.15 Release 2.1.2 (2018-11-06)
• added async_eigen_decomp parameter for ACKTR and set it to False by default (remove deprecation
warnings)
• added methods for calling env methods/setting attributes inside a VecEnv (thanks to @bjmuld)
• updated gym minimum version
1.38.16 Release 2.1.1 (2018-10-20)
• fixed MpiAdam synchronization issue in PPO1 (thanks to @brendenpetersen) issue #50

• fixed dependency issues (new mujoco-py requires a mujoco license + gym broke MultiDiscrete space shape)
1.38.17 Release 2.1.0 (2018-10-2)
Warning: This version contains breaking changes for DQN policies, please read the full details
Bug fixes + doc update

• added patch fix for equal function using gym.spaces.MultiDiscrete and gym.spaces.MultiBinary
• fixes for DQN action_probability
• re-added double DQN + refactored DQN policies breaking changes
• replaced async with async_eigen_decomp in ACKTR/KFAC for python 3.7 compatibility
• removed action clipping for prediction of continuous actions (see issue #36)
• fixed NaN issue due to clipping the continuous action in the wrong place (issue #36)
• documentation was updated (policy + DDPG example hyperparameters)

1.38.18 Release 2.0.0 (2018-09-18)
Warning: This version contains breaking changes, please read the full details
Tensorboard, refactoring and bug fixes

• Renamed DeepQ to DQN breaking changes
• Renamed DeepQPolicy to DQNPolicy breaking changes
• fixed DDPG behavior breaking changes
• changed default policies for DDPG, so that DDPG now works correctly breaking changes
• added more documentation (some modules from common).
• added doc about using custom env
• added Tensorboard support for A2C, ACER, ACKTR, DDPG, DeepQ, PPO1, PPO2 and TRPO
• added episode reward to Tensorboard
• added documentation for Tensorboard usage
• added Identity for Box action space
• fixed render function ignoring parameters when using wrapped environments
• fixed PPO1 and TRPO done values for recurrent policies
• fixed image normalization not occurring when using images
• updated VecEnv objects for the new Gym version
• added test for DDPG
• refactored DQN policies
• added registry for policies, can be passed as string to the agent
• added documentation for custom policies + policy registration
• fixed numpy warning when using DDPG Memory
• fixed DummyVecEnv not copying the observation array when stepping and resetting
• added pre-built docker images + installation instructions
• added deterministic argument in the predict function
• added assert in PPO2 for recurrent policies
• fixed predict function to handle both vectorized and unwrapped environment
• added input check to the predict function
• refactored ActorCritic models to reduce code duplication
• refactored Off Policy models (to begin HER and replay_buffer refactoring)
• added tests for auto vectorization detection
• fixed render function, to handle positional arguments
1.38. Changelog 205

1.38.19 Release 1.0.7 (2018-08-29)
Bug fixes and documentation

• added html documentation using sphinx + integration with read the docs
• cleaned up README + typos
• fixed normalization for DQN with images
• fixed DQN identity test
1.38.20 Release 1.0.1 (2018-08-20)
Refactored Stable Baselines

• refactored A2C, ACER, ACTKR, DDPG, DeepQ, GAIL, TRPO, PPO1 and PPO2 under a single constant class
• added callback to refactored algorithm training
• added saving and loading to refactored algorithms
• refactored ACER, DDPG, GAIL, PPO1 and TRPO to fit with A2C, PPO2 and ACKTR policies
• added new policies for most algorithms (Mlp, MlpLstm, MlpLnLstm, Cnn, CnnLstm and CnnLnLstm)
• added dynamic environment switching (so continual RL learning is now feasible)
• added prediction from observation and action probability from observation for all the algorithms
• fixed graphs issues, so models wont collide in names
• fixed behavior_clone weight loading for GAIL
• fixed Tensorflow using all the GPU VRAM
• fixed models so that they are all compatible with vectorized environments
• fixed set_global_seed to update gym.spaces’s random seed
• fixed PPO1 and TRPO performance issues when learning identity function
• added new tests for loading, saving, continuous actions and learning the identity function
• fixed DQN wrapping for atari
• added saving and loading for Vecnormalize wrapper
• added automatic detection of action space (for the policy network)
• fixed ACER buffer with constant values assuming n_stack=4
• fixed some RL algorithms not clipping the action to be in the action_space, when using gym.spaces.Box
• refactored algorithms can take either a gym.Environment or a str ([if the environment name is regis-
tered](https://github.com/openai/gym/wiki/Environments))
• Hoftix in ACER (compared to v1.0.0)
Future Work :
• Finish refactoring HER
• Refactor ACKTR and ACER for continuous implementation

1.38.21 Release 0.1.6 (2018-07-27)
Deobfuscation of the code base + pep8 and fixes

• Fixed tf.session().__enter__() being used, rather than sess = tf.session() and passing the
session to the objects
• Fixed uneven scoping of TensorFlow Sessions throughout the code
• Fixed rolling vecwrapper to handle observations that are not only grayscale images
• Fixed deepq saving the environment when trying to save itself
• Fixed ValueError: Cannot take the length of Shape with unknown rank. in
acktr, when running run_atari.py script.
• Fixed calling baselines sequentially no longer creates graph conflicts
• Fixed mean on empty array warning with deepq
• Fixed kfac eigen decomposition not cast to float64, when the parameter use_float64 is set to True
• Fixed Dataset data loader, not correctly resetting id position if shuffling is disabled
• Fixed EOFError when reading from connection in the worker in subproc_vec_env.py
• Fixed behavior_clone weight loading and saving for GAIL
• Avoid taking root square of negative number in trpo_mpi.py
• Removed some duplicated code (a2cpolicy, trpo_mpi)
• Removed unused, undocumented and crashing function reset_task in subproc_vec_env.py
• Reformated code to PEP8 style
• Documented all the codebase
• Added atari tests
• Added logger tests
Missing: tests for acktr continuous (+ HER, rely on mujoco. . . )
1.38.22 Maintainers
Stable-Baselines is currently maintained by Ashley Hill (aka @hill-a), Antonin Raffin (aka @araffin), Maximilian
Ernestus (aka @ernestum), Adam Gleave (@AdamGleave) and Anssi Kanervisto (aka @Miffyli).
1.38.23 Contributors (since v2.0.0):
In random order. . .
Thanks to @bjmuld @iambenzo @iandanforth @r7vme @brendenpetersen @huvar @abhiskk @JohannesAck
@mily20001 @EliasHasle @mrakgr @Bleyddyn @antoine-galataud @junhyeokahn @AdamGleave @keshaviyen-
gar @tperol @XMaster96 @kantneel @Pastafarianist @GerardMaggiolino @PatrickWalter214 @yutingsz @sc420
@Aaahh @billtubbs @Miffyli @dwiel @miguelrass @qxcv @jaberkow @eavelardev @ruifeng96150 @pedrohbtp
@srivatsankrishnan @evilsocket @MarvineGothic @jdossgollin @SyllogismRXS @rusu24edward @jbulow @An-
tymon @seheevic @justinkterry @edbeeching @flodorner @KuKuXia @NeoExtended @PartiallyTyped @mmcenta
@richardwu @tirafesi @caburu @johannes-dornheim @kvenkman @aakash94 @enderdead @hardmaru @jbarsce
@ColinLeongUDRI @shwang @YangRui2015 @sophiagu @OGordon100 @SVJayanthi @sunshineclt @roccivic
@anj1
1.38. Changelog 207

1.39 Projects
This is a list of projects using stable-baselines. Please tell us, if you want your project to appear on this page ;)
1.39.1 Stable Baselines for TensorFlow 2
A fork of the original stable-baselines repo that works with TF2.x.
Author: Sophia Gu (@sophiagu)

Github repo: https://github.com/sophiagu/stable-baselines-tf2
1.39.2 Slime Volleyball Gym Environment
A simple environment for benchmarking single and multi-agent reinforcement learning algorithms on a clone of the
Slime Volleyball game. Only dependencies are gym and numpy. Both state and pixel observation environments are
available. The motivation of this environment is to easily enable trained agents to play against each other, and also
facilitate the training of agents directly in a multi-agent setting, thus adding an extra dimension for evaluating an
agent’s performance.
Uses stable-baselines to train RL agents for both state and pixel observation versions of the task. A tutorial is also
provided on modifying stable-baselines for self-play using PPO.
Author: David Ha (@hardmaru)

Github repo: https://github.com/hardmaru/slimevolleygym
1.39.3 Learning to drive in a day
Implementation of reinforcement learning approach to make a donkey car learn to drive. Uses DDPG on VAE features
(reproducing paper from wayve.ai)
Author: Roma Sokolkov (@r7vme)

Github repo: https://github.com/r7vme/learning-to-drive-in-a-day
1.39.4 Donkey Gym
OpenAI gym environment for donkeycar simulator.
Author: Tawn Kramer (@tawnkramer)

Github repo: https://github.com/tawnkramer/donkey_gym
1.39.5 Self-driving FZERO Artificial Intelligence
Series of videos on how to make a self-driving FZERO artificial intelligence using reinforcement learning algorithms
PPO2 and A2C.

Author: Lucas Thompson

Video Link
1.39.6 S-RL Toolbox
S-RL Toolbox: Reinforcement Learning (RL) and State Representation Learning (SRL) for Robotics. Stable-Baselines
was originally developped for this project.
Authors: Antonin Raffin, Ashley Hill, René Traoré, Timothée Lesort, Natalia Díaz-Rodríguez, David Filliat
Github repo: https://github.com/araffin/robotics-rl-srl
1.39.7 Roboschool simulations training on Amazon SageMaker
“In this notebook example, we will make HalfCheetah learn to walk using the stable-baselines [. . . ]”
Author: Amazon AWS

Repo Link
1.39.8 MarathonEnvs + OpenAi.Baselines
Experimental - using OpenAI baselines with MarathonEnvs (ML-Agents)
Author: Joe Booth (@Sohojoe)

Github repo: https://github.com/Sohojoe/MarathonEnvsBaselines
1.39.9 Learning to drive smoothly in minutes
Implementation of reinforcement learning approach to make a car learn to drive smoothly in minutes. Uses SAC on
VAE features.
Author: Antonin Raffin (@araffin)

Blog post: https://towardsdatascience.com/learning-to-drive-smoothly-in-minutes-450a7cdb35f4
Github repo: https://github.com/araffin/learning-to-drive-in-5-minutes
1.39.10 Making Roboy move with elegance
Project around Roboy, a tendon-driven robot, that enabled it to move its shoulder in simulation to reach a pre-defined
point in 3D space. The agent used Proximal Policy Optimization (PPO) or Soft Actor-Critic (SAC) and was tested on
the real hardware.
Authors: Alexander Pakakis, Baris Yazici, Tomas Ruiz

Email: [email protected]
1.39. Projects 209

GitHub repo: https://github.com/Roboy/DeepAndReinforced

DockerHub image: deepandreinforced/rl:latest
Presentation: https://tinyurl.com/DeepRoboyControl
Video: https://tinyurl.com/DeepRoboyControlVideo
Blog post: https://tinyurl.com/mediumDRC
Website: https://roboy.org/
1.39.11 Train a ROS-integrated mobile robot (differential drive) to avoid dynamic

objects
The RL-agent serves as local planner and is trained in a simulator, fusion of the Flatland Simulator and the crowd
simulator Pedsim. This was tested on a real mobile robot. The Proximal Policy Optimization (PPO) algorithm is
applied.
Author: Ronja Güldenring

Video: https://www.youtube.com/watch?v=laGrLaMaeT4
GitHub: https://github.com/RGring/drl_local_planner_ros_stable_baselines
1.39.12 Adversarial Policies: Attacking Deep Reinforcement Learning
Uses Stable Baselines to train adversarial policies that attack pre-trained victim policies in a zero-sum multi-agent
environments. May be useful as an example of how to integrate Stable Baselines with Ray to perform distributed
experiments and Sacred for experiment configuration and monitoring.
Authors: Adam Gleave, Michael Dennis, Neel Kant, Cody Wild

GitHub: https://github.com/HumanCompatibleAI/adversarial-policies
Paper: https://arxiv.org/abs/1905.10615
Website: https://adversarialpolicies.github.io
1.39.13 WaveRL: Training RL agents to perform active damping
Reinforcement learning is used to train agents to control pistons attached to a bridge to cancel out vibrations. The
bridge is modeled as a one dimensional oscillating system and dynamics are simulated using a finite difference solver.
Agents were trained using Proximal Policy Optimization. See presentation for environment detalis.
Author: Jack Berkowitz

GitHub: https://github.com/jaberkow/WaveRL
Presentation: http://bit.ly/WaveRLslides

1.39.14 Fenics-DRL: Fluid mechanics and Deep Reinforcement Learning
Deep Reinforcement Learning is used to control the position or the shape of obstacles in different fluids in order to
optimize drag or lift. Fenics is used for the Fluid Mechanics part, and Stable Baselines is used for the DRL.
Authors: Paul Garnier, Jonathan Viquerat, Aurélien Larcher, Elie Hachem

GitHub: https://github.com/DonsetPG/openFluid
Website: https://donsetpg.github.io/blog/2019/08/06/DRL-FM-review/
1.39.15 Air Learning: An AI Research Platform Algorithm Hardware Benchmarking

of Autonomous Aerial Robots
Aerial robotics is a cross-layer, interdisciplinary field. Air Learning is an effort to bridge seemingly disparate fields.
Designing an autonomous robot to perform a task involves interactions between various boundaries spanning from
modeling the environment down to the choice of onboard computer platform available in the robot. Our goal through
building Air Learning is to provide researchers with a cross-domain infrastructure that allows them to holistically study
and evaluate reinforcement learning algorithms for autonomous aerial machines. We use stable-baselines to train UAV
agent with Deep Q-Networks and Proximal Policy Optimization algorithms.
Authors: Srivatsan Krishnan, Behzad Boroujerdian, William Fu, Aleksandra Faust, Vijay Janapa Reddi
Github: https://github.com/harvard-edge/airlearning
Paper: https://arxiv.org/pdf/1906.00421.pdf
Video: https://www.youtube.com/watch?v=oakzGnh7Llw (Simulation),
https://www.youtube.com/watch?v=cvO5YOzI0mg (on a CrazyFlie Nano-Drone)
1.39.16 Snake Game AI
AI to play the classic snake game. The game was trained using PPO2 available from stable-baselines and then exported
to tensorflowjs to run directly on the browser
Author: Pedro Torres (@pedrohbtp)

Repository: https://github.com/pedrohbtp/snake-rl
Website: https://www.pedro-torres.com/snake-rl/
1.39.17 Pwnagotchi
Pwnagotchi is an A2C-based “AI” powered by bettercap and running on a Raspberry Pi Zero W that learns from its
surrounding WiFi environment in order to maximize the crackable WPA key material it captures (either through passive
sniffing or by performing deauthentication and association attacks). This material is collected on disk as PCAP files
containing any form of handshake supported by hashcat, including full and half WPA handshakes as well as PMKIDs.
1.39. Projects 211

Author: Simone Margaritelli (@evilsocket)

Repository: https://github.com/evilsocket/pwnagotchi
Website: https://pwnagotchi.ai/
1.39.18 Quantized Reinforcement Learning (QuaRL)
QuaRL is a open-source framework to study the effects of quantization broad spectrum of reinforcement learning
algorithms. The RL algorithms we used in this study are from stable-baselines.
Authors: Srivatsan Krishnan, Sharad Chitlangia, Maximilian Lam, Zishen Wan, Aleksandra Faust, Vijay Janapa
Reddi
Github: https://github.com/harvard-edge/quarl
Paper: https://arxiv.org/pdf/1910.01055.pdf
1.39.19 PPO_CPP: C++ version of a Deep Reinforcement Learning algorithm PPO
Executes PPO at C++ level yielding notable execution performance speedups. Uses Stable Baselines to create a
computational graph which is then used for training with custom environments by machine-code-compiled binary.
Author: Szymon Brych

GitHub: https://github.com/Antymon/ppo_cpp
1.39.20 Learning Agile Robotic Locomotion Skills by Imitating Animals
Learning locomotion gaits by imitating animals. It uses PPO1 and AWR.
Authors: Xue Bin Peng, Erwin Coumans, Tingnan Zhang, Tsang-Wei Lee, Jie Tan, Sergey Levine
Website: https://xbpeng.github.io/projects/Robotic_Imitation/index.html
Github: https://github.com/google-research/motion_imitation
1.39.21 Imitation Learning Baseline Implementations
This project aims to provide clean implementations of imitation learning algorithms. Currently we have implementa-
tions of AIRL and GAIL, and intend to add more in the future.
Authors: Adam Gleave, Steven Wang, Nevan Wichers, Sam Toyer

Github: https://github.com/HumanCompatibleAI/imitation

1.40 Plotting Results
stable_baselines.results_plotter.main()
Example usage in jupyter-notebook
from stable_baselines import results_plotter

%matplotlib inline
results_plotter.plot_results(["./log"], 10e6, results_plotter.X_TIMESTEPS,
˓→"Breakout")
Here ./log is a directory containing the monitor.csv files

stable_baselines.results_plotter.plot_curves(xy_list, xaxis, title)
plot the curves
Parameters
• xy_list – ([(np.ndarray, np.ndarray)]) the x and y coordinates to plot
• xaxis – (str) the axis for the x and y output (can be X_TIMESTEPS=’timesteps’,
X_EPISODES=’episodes’ or X_WALLTIME=’walltime_hrs’)
• title – (str) the title of the plot
stable_baselines.results_plotter.plot_results(dirs, num_timesteps, xaxis, task_name)
plot the results
Parameters
• dirs – ([str]) the save location of the results to plot
• num_timesteps – (int or None) only plot the points below this value
• task_name – (str) the title of the task to plot
stable_baselines.results_plotter.rolling_window(array, window)
apply a rolling window to a np.ndarray
Parameters
• array – (np.ndarray) the input Array
• window – (int) length of the rolling window
Returns (np.ndarray) rolling window on the input array
stable_baselines.results_plotter.ts2xy(timesteps, xaxis)
Decompose a timesteps variable to x ans ys
Parameters
• timesteps – (Pandas DataFrame) the input data
Returns (np.ndarray, np.ndarray) the x and y output
stable_baselines.results_plotter.window_func(var_1, var_2, window, func)
apply a function to the rolling window of 2 arrays
Parameters
1.40. Plotting Results 213

• var_1 – (np.ndarray) variable 1

• var_2 – (np.ndarray) variable 2
• window – (int) length of the rolling window
• func – (numpy function) function to apply on the rolling window on variable 2 (such as
np.mean)
Returns (np.ndarray, np.ndarray) the rolling output with applied function

CHAPTER 2
Citing Stable Baselines
To cite this project in publications:
@misc{stable-baselines,
author = {Hill, Ashley and Raffin, Antonin and Ernestus, Maximilian and Gleave,
˓→Adam and Kanervisto, Anssi and Traore, Rene and Dhariwal, Prafulla and Hesse,
˓→Christopher and Klimov, Oleg and Nichol, Alex and Plappert, Matthias and Radford,
˓→Alec and Schulman, John and Sidor, Szymon and Wu, Yuhuai},
title = {Stable Baselines},

year = {2018},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/hill-a/stable-baselines}},
}
215
216 Chapter 2. Citing Stable Baselines

CHAPTER 3
Contributing
To any interested in making the rl baselines better, there are still some improvements that need to be done. A full
TODO list is available in the roadmap.
If you want to contribute, please read CONTRIBUTING.md first.
217
218 Chapter 3. Contributing

CHAPTER 4
Indices and tables
• genindex
• search
• modindex
219
220 Chapter 4. Indices and tables

Python Module Index
s
stable_baselines.a2c, 71
stable_baselines.acer, 77
stable_baselines.acktr, 83
stable_baselines.bench.monitor, 190
stable_baselines.common.base_class, 61
stable_baselines.common.callbacks, 42
stable_baselines.common.cmd_util, 185
stable_baselines.common.distributions,
174
stable_baselines.common.env_checker, 189
stable_baselines.common.evaluation, 189
stable_baselines.common.policies, 64
stable_baselines.common.schedules, 187
stable_baselines.common.tf_util, 181
stable_baselines.common.vec_env, 25
stable_baselines.ddpg, 89
stable_baselines.deepq, 106
stable_baselines.gail, 118
stable_baselines.her, 124
stable_baselines.ppo1, 128
stable_baselines.ppo2, 134
stable_baselines.results_plotter, 213
stable_baselines.sac, 140
stable_baselines.td3, 154
stable_baselines.trpo_mpi, 167
221
222 Python Module Index

Index
A 160
A2C (class in stable_baselines.a2c), 73 action_probability() (stable_baselines.a2c.A2C
ACER (class in stable_baselines.acer), 79 method), 74
ACKTR (class in stable_baselines.acktr), 85 action_probability() (sta-
action (stable_baselines.common.policies.ActorCriticPolicy ble_baselines.acer.ACER method), 80
attribute), 66 action_probability() (sta-
action_ph (stable_baselines.common.policies.BasePolicy ble_baselines.acktr.ACKTR method), 86
attribute), 65 action_probability() (sta-
action_ph (stable_baselines.ddpg.CnnPolicy at- ble_baselines.common.base_class.BaseRLModel
tribute), 99 method), 61
action_ph (stable_baselines.ddpg.LnCnnPolicy action_probability() (sta-
attribute), 101 ble_baselines.ddpg.DDPG method), 93
action_ph (stable_baselines.ddpg.LnMlpPolicy action_probability() (sta-
attribute), 98 ble_baselines.deepq.DQN method), 109
action_ph (stable_baselines.ddpg.MlpPolicy at- action_probability() (sta-
tribute), 96 ble_baselines.gail.GAIL method), 121
action_ph (stable_baselines.deepq.CnnPolicy at- action_probability() (stable_baselines.her.HER
action_ph (stable_baselines.deepq.LnCnnPolicy at- action_probability() (sta-
tribute), 115 ble_baselines.ppo1.PPO1 method), 131
action_ph (stable_baselines.deepq.LnMlpPolicy at- action_probability() (sta-
tribute), 113 ble_baselines.ppo2.PPO2 method), 137
action_ph (stable_baselines.deepq.MlpPolicy at- action_probability() (stable_baselines.sac.SAC
action_ph (stable_baselines.sac.CnnPolicy attribute), action_probability() (stable_baselines.td3.TD3
150 method), 157
action_ph (stable_baselines.sac.LnCnnPolicy at- action_probability() (sta-
tribute), 151 ble_baselines.trpo_mpi.TRPO method),
action_ph (stable_baselines.sac.LnMlpPolicy at- 169
tribute), 148 ActorCriticPolicy (class in sta-
action_ph (stable_baselines.sac.MlpPolicy attribute), ble_baselines.common.policies), 66
146 adapt() (stable_baselines.ddpg.AdaptiveParamNoiseSpec
action_ph (stable_baselines.td3.CnnPolicy attribute), method), 103
163 AdaptiveParamNoiseSpec (class in sta-
action_ph (stable_baselines.td3.LnCnnPolicy at- ble_baselines.ddpg), 102
tribute), 165 add() (stable_baselines.her.HindsightExperienceReplayWrapper
action_ph (stable_baselines.td3.LnMlpPolicy at- method), 128
tribute), 162 arg_parser() (in module sta-
action_ph (stable_baselines.td3.MlpPolicy attribute), ble_baselines.common.cmd_util), 185
atari_arg_parser() (in module sta-
223
ble_baselines.common.cmd_util), 185 CnnPolicy (class in sta-

avg_norm() (in module sta- ble_baselines.common.policies), 70
ble_baselines.common.tf_util), 181 CnnPolicy (class in stable_baselines.ddpg), 99
CnnPolicy (class in stable_baselines.deepq), 114
B CnnPolicy (class in stable_baselines.sac), 149
BaseCallback (class in sta- CnnPolicy (class in stable_baselines.td3), 163
ble_baselines.common.callbacks), 43 constant() (in module sta-
BasePolicy (class in sta- ble_baselines.common.schedules), 188
ble_baselines.common.policies), 65 ConstantSchedule (class in sta-
BaseRLModel (class in sta- ble_baselines.common.schedules), 187
ble_baselines.common.base_class), 61 constfn() (in module sta-
batch_to_seq() (in module sta- ble_baselines.common.schedules), 188
ble_baselines.common.tf_util), 181 convert_dict_to_obs() (sta-
BernoulliProbabilityDistribution (class in ble_baselines.her.HERGoalEnvWrapper
stable_baselines.common.distributions), 174 method), 127
BernoulliProbabilityDistributionType convert_obs_to_dict() (sta-
(class in sta- ble_baselines.her.HERGoalEnvWrapper
ble_baselines.common.distributions), 174 method), 128
ConvertCallback (class in sta-
C ble_baselines.common.callbacks), 43
calc_entropy() (in module sta-
ble_baselines.common.tf_util), 181
D
CallbackList (class in sta- DataLoader (class in stable_baselines.gail), 52
ble_baselines.common.callbacks), 43 DDPG (class in stable_baselines.ddpg), 91
deterministic_action
can_sample() (stable_baselines.her.HindsightExperienceReplayWrapper (sta-
method), 128 ble_baselines.common.policies.ActorCriticPolicy
CategoricalProbabilityDistribution (class attribute), 66
in stable_baselines.common.distributions), 175 DiagGaussianProbabilityDistribution
CategoricalProbabilityDistributionType (class in sta-
(class in sta- ble_baselines.common.distributions), 176
ble_baselines.common.distributions), 175 DiagGaussianProbabilityDistributionType
check_env() (in module sta- (class in sta-
ble_baselines.common.env_checker), 189 ble_baselines.common.distributions), 177
check_shape() (in module sta- double_linear_con() (in module sta-
ble_baselines.common.tf_util), 181 ble_baselines.common.schedules), 188
CheckpointCallback (class in sta- double_middle_drop() (in module sta-
ble_baselines.common.callbacks), 43 ble_baselines.common.schedules), 188
close() (stable_baselines.bench.monitor.Monitor DQN (class in stable_baselines.deepq), 108
method), 190 DummyVecEnv (class in sta-
close() (stable_baselines.common.vec_env.DummyVecEnv ble_baselines.common.vec_env), 27
method), 27
close() (stable_baselines.common.vec_env.SubprocVecEnv E
method), 29 entropy() (stable_baselines.common.distributions.BernoulliProbabilityD
close() (stable_baselines.common.vec_env.VecEnv method), 174
method), 26 entropy() (stable_baselines.common.distributions.CategoricalProbabili
close() (stable_baselines.common.vec_env.VecFrameStack method), 175
method), 30 entropy() (stable_baselines.common.distributions.DiagGaussianProbab
close() (stable_baselines.common.vec_env.VecVideoRecorder method), 176
method), 31 entropy() (stable_baselines.common.distributions.MultiCategoricalProb
CnnLnLstmPolicy (class in sta- method), 177
ble_baselines.common.policies), 71 entropy() (stable_baselines.common.distributions.ProbabilityDistributio
CnnLstmPolicy (class in sta- method), 179
ble_baselines.common.policies), 71 env_method() (stable_baselines.common.vec_env.DummyVecEnv
method), 27
224 Index
env_method() (stable_baselines.common.vec_env.SubprocVecEnv
get_env() (stable_baselines.common.base_class.BaseRLModel
method), 29 method), 62
env_method() (stable_baselines.common.vec_env.VecEnvget_env() (stable_baselines.ddpg.DDPG method), 93
method), 26 get_env() (stable_baselines.deepq.DQN method), 109
EvalCallback (class in sta- get_env() (stable_baselines.gail.GAIL method), 122
ble_baselines.common.callbacks), 43 get_env() (stable_baselines.her.HER method), 126
evaluate_policy() (in module sta- get_env() (stable_baselines.ppo1.PPO1 method), 131
ble_baselines.common.evaluation), 189 get_env() (stable_baselines.ppo2.PPO2 method), 137
EventCallback (class in sta- get_env() (stable_baselines.sac.SAC method), 144
ble_baselines.common.callbacks), 44 get_env() (stable_baselines.td3.TD3 method), 158
EveryNTimesteps (class in sta- get_env() (stable_baselines.trpo_mpi.TRPO method),
ble_baselines.common.callbacks), 44 170
ExpertDataset (class in stable_baselines.gail), 51 get_episode_lengths() (sta-
ble_baselines.bench.monitor.Monitor method),
F 190
FeedForwardPolicy (class in sta- get_episode_rewards() (sta-
ble_baselines.common.policies), 67 ble_baselines.bench.monitor.Monitor method),
flatgrad() (in module sta- 190
ble_baselines.common.tf_util), 181 get_episode_times() (sta-
ble_baselines.bench.monitor.Monitor method),
flatparam() (stable_baselines.common.distributions.BernoulliProbabilityDistribution
method), 174 190
get_globals_vars()
flatparam() (stable_baselines.common.distributions.CategoricalProbabilityDistribution(in module sta-
method), 175 ble_baselines.common.tf_util), 182
get_images() (stable_baselines.common.vec_env.DummyVecEnv
flatparam() (stable_baselines.common.distributions.DiagGaussianProbabilityDistribution
get_images() (stable_baselines.common.vec_env.SubprocVecEnv
flatparam() (stable_baselines.common.distributions.MultiCategoricalProbabilityDistribution
get_images() (stable_baselines.common.vec_env.VecEnv
flatparam() (stable_baselines.common.distributions.ProbabilityDistribution
get_monitor_files()
fromflat() (stable_baselines.common.distributions.BernoulliProbabilityDistribution (in module sta-
class method), 174 ble_baselines.bench.monitor), 191
get_next_batch()
fromflat() (stable_baselines.common.distributions.CategoricalProbabilityDistribution (sta-
class method), 175 ble_baselines.gail.ExpertDataset method),
52
fromflat() (stable_baselines.common.distributions.DiagGaussianProbabilityDistribution
class method), 176 get_original_obs() (sta-
ble_baselines.common.vec_env.VecNormalize
fromflat() (stable_baselines.common.distributions.MultiCategoricalProbabilityDistribution
class method), 178 method), 30
function() (in module sta- get_original_reward() (sta-
ble_baselines.common.tf_util), 181 ble_baselines.common.vec_env.VecNormalize
method), 30
G get_parameter_list() (stable_baselines.a2c.A2C
GAIL (class in stable_baselines.gail), 120 method), 74
generate_expert_traj() (in module sta- get_parameter_list() (sta-
ble_baselines.gail), 53 ble_baselines.acer.ACER method), 80
get_attr() (stable_baselines.common.vec_env.DummyVecEnv (sta-
method), 27 ble_baselines.acktr.ACKTR method), 86
get_attr() (stable_baselines.common.vec_env.SubprocVecEnv (sta-
method), 29 ble_baselines.common.base_class.BaseRLModel
get_attr() (stable_baselines.common.vec_env.VecEnv method), 62
method), 26 get_parameter_list() (sta-
get_env() (stable_baselines.a2c.A2C method), 74 ble_baselines.ddpg.DDPG method), 93
get_env() (stable_baselines.acer.ACER method), 80 get_parameter_list() (sta-
get_env() (stable_baselines.acktr.ACKTR method), 86 ble_baselines.deepq.DQN method), 109
get_parameter_list() (sta-
Index 225
ble_baselines.gail.GAIL method), 122 ble_baselines.acktr.ACKTR method), 86

get_parameter_list() (stable_baselines.her.HER get_vec_normalize_env() (sta-
method), 126 ble_baselines.common.base_class.BaseRLModel
get_parameter_list() (sta- method), 62
ble_baselines.ppo1.PPO1 method), 131 get_vec_normalize_env() (sta-
get_parameter_list() (sta- ble_baselines.ddpg.DDPG method), 93
ble_baselines.ppo2.PPO2 method), 137 get_vec_normalize_env() (sta-
get_parameter_list() (stable_baselines.sac.SAC ble_baselines.deepq.DQN method), 110
method), 144 get_vec_normalize_env() (sta-
get_parameter_list() (stable_baselines.td3.TD3 ble_baselines.gail.GAIL method), 122
method), 158 get_vec_normalize_env() (sta-
get_parameter_list() (sta- ble_baselines.ppo1.PPO1 method), 132
ble_baselines.trpo_mpi.TRPO method), get_vec_normalize_env() (sta-
170 ble_baselines.ppo2.PPO2 method), 138
get_parameters() (stable_baselines.a2c.A2C get_vec_normalize_env() (sta-
method), 74 ble_baselines.sac.SAC method), 144
get_parameters() (stable_baselines.acer.ACER get_vec_normalize_env() (sta-
method), 80 ble_baselines.td3.TD3 method), 158
get_parameters() (stable_baselines.acktr.ACKTR get_vec_normalize_env() (sta-
method), 86 ble_baselines.trpo_mpi.TRPO method),
get_parameters() (sta- 170
ble_baselines.common.base_class.BaseRLModel getattr_depth_check() (sta-
method), 62 ble_baselines.common.vec_env.VecEnv
get_parameters() (stable_baselines.ddpg.DDPG method), 26
method), 93 GoalSelectionStrategy (class in sta-
get_parameters() (stable_baselines.deepq.DQN ble_baselines.her), 127
method), 109 gradient_add() (in module sta-
get_parameters() (stable_baselines.gail.GAIL ble_baselines.common.tf_util), 182
method), 122
get_parameters() (stable_baselines.ppo1.PPO1 H
method), 131 HER (class in stable_baselines.her), 125
get_parameters() (stable_baselines.ppo2.PPO2 HERGoalEnvWrapper (class in stable_baselines.her),
method), 138 127
get_parameters() (stable_baselines.sac.SAC HindsightExperienceReplayWrapper (class in
method), 144 stable_baselines.her), 128
get_parameters() (stable_baselines.td3.TD3 huber_loss() (in module sta-
method), 158 ble_baselines.common.tf_util), 182
get_parameters() (sta-
ble_baselines.trpo_mpi.TRPO method), I
170 in_session() (in module sta-
get_schedule_fn() (in module sta- ble_baselines.common.tf_util), 183
ble_baselines.common.schedules), 188 init_callback() (sta-
get_stats() (stable_baselines.ddpg.AdaptiveParamNoiseSpec ble_baselines.common.callbacks.BaseCallback
get_total_steps() (sta- init_callback() (sta-
ble_baselines.bench.monitor.Monitor method), ble_baselines.common.callbacks.EventCallback
190 method), 44
get_trainable_vars() (in module sta- init_dataloader() (sta-
ble_baselines.common.tf_util), 182 ble_baselines.gail.ExpertDataset method),
get_vec_normalize_env() (sta- 52
ble_baselines.a2c.A2C method), 74 initial_state (sta-
get_vec_normalize_env() (sta- ble_baselines.common.policies.BasePolicy
ble_baselines.acer.ACER method), 80 attribute), 65
get_vec_normalize_env() (sta-
226 Index
initial_state (stable_baselines.ddpg.CnnPolicy attribute), 112

tribute), 99 is_discrete (stable_baselines.sac.CnnPolicy at-
initial_state (stable_baselines.ddpg.LnCnnPolicy tribute), 150
attribute), 101 is_discrete (stable_baselines.sac.LnCnnPolicy at-
initial_state (stable_baselines.ddpg.LnMlpPolicy tribute), 151
attribute), 98 is_discrete (stable_baselines.sac.LnMlpPolicy at-
initial_state (stable_baselines.ddpg.MlpPolicy attribute), 148
tribute), 96 is_discrete (stable_baselines.sac.MlpPolicy at-
initial_state (stable_baselines.deepq.CnnPolicy tribute), 147
attribute), 114 is_discrete (stable_baselines.td3.CnnPolicy at-
initial_state (sta- tribute), 163
ble_baselines.deepq.LnCnnPolicy attribute), is_discrete (stable_baselines.td3.LnCnnPolicy at-
116 tribute), 165
initial_state (stable_baselines.deepq.LnMlpPolicy is_discrete (stable_baselines.td3.LnMlpPolicy at-
attribute), 113 tribute), 162
initial_state (stable_baselines.deepq.MlpPolicy is_discrete (stable_baselines.td3.MlpPolicy at-
attribute), 112 tribute), 161
initial_state (stable_baselines.sac.CnnPolicy at- is_image() (in module sta-
tribute), 150 ble_baselines.common.tf_util), 183
initial_state (stable_baselines.sac.LnCnnPolicy is_using_her() (stable_baselines.ddpg.DDPG
attribute), 151 method), 93
initial_state (stable_baselines.sac.LnMlpPolicy is_using_her() (stable_baselines.deepq.DQN
initial_state (stable_baselines.sac.MlpPolicy at- is_using_her() (stable_baselines.sac.SAC method),
tribute), 147 144
initial_state (stable_baselines.td3.CnnPolicy at- is_using_her() (stable_baselines.td3.TD3 method),
tribute), 163 158
initial_state (stable_baselines.td3.LnCnnPolicy
attribute), 165 K
initial_state (stable_baselines.td3.LnMlpPolicy kl() (stable_baselines.common.distributions.BernoulliProbabilityDistribu
initial_state (stable_baselines.td3.MlpPolicy at- kl() (stable_baselines.common.distributions.CategoricalProbabilityDistri
initialize() (in module sta- kl() (stable_baselines.common.distributions.DiagGaussianProbabilityDis
ble_baselines.common.tf_util), 183 method), 176
intprod() (in module sta- kl() (stable_baselines.common.distributions.MultiCategoricalProbabilityD
ble_baselines.common.tf_util), 183 method), 178
is_discrete (stable_baselines.common.policies.BasePolicy
kl() (stable_baselines.common.distributions.ProbabilityDistribution
is_discrete (stable_baselines.ddpg.CnnPolicy at-
tribute), 99 L
is_discrete (stable_baselines.ddpg.LnCnnPolicy at- learn() (stable_baselines.a2c.A2C method), 74
tribute), 101 learn() (stable_baselines.acer.ACER method), 81
is_discrete (stable_baselines.ddpg.LnMlpPolicy at- learn() (stable_baselines.acktr.ACKTR method), 86
tribute), 98 learn() (stable_baselines.common.base_class.BaseRLModel
is_discrete (stable_baselines.ddpg.MlpPolicy at- method), 62
tribute), 96 learn() (stable_baselines.ddpg.DDPG method), 94
is_discrete (stable_baselines.deepq.CnnPolicy at- learn() (stable_baselines.deepq.DQN method), 110
tribute), 114 learn() (stable_baselines.gail.GAIL method), 122
is_discrete (stable_baselines.deepq.LnCnnPolicy learn() (stable_baselines.her.HER method), 126
attribute), 116 learn() (stable_baselines.ppo1.PPO1 method), 132
is_discrete (stable_baselines.deepq.LnMlpPolicy learn() (stable_baselines.ppo2.PPO2 method), 138
attribute), 113 learn() (stable_baselines.sac.SAC method), 144
is_discrete (stable_baselines.deepq.MlpPolicy at- learn() (stable_baselines.td3.TD3 method), 158
Index 227
learn() (stable_baselines.trpo_mpi.TRPO method), method), 132

170 load_parameters() (stable_baselines.ppo2.PPO2
linear_interpolation() (in module sta- method), 138
ble_baselines.common.schedules), 188 load_parameters() (stable_baselines.sac.SAC
linear_schedule() (in module sta- method), 145
ble_baselines.common.schedules), 188 load_parameters() (stable_baselines.td3.TD3
LinearSchedule (class in sta- method), 159
ble_baselines.common.schedules), 187 load_parameters() (sta-
LnCnnPolicy (class in stable_baselines.ddpg), 101 ble_baselines.trpo_mpi.TRPO method),
LnCnnPolicy (class in stable_baselines.deepq), 115 171
LnCnnPolicy (class in stable_baselines.sac), 151 load_results() (in module sta-
LnCnnPolicy (class in stable_baselines.td3), 164 ble_baselines.bench.monitor), 191
LnMlpPolicy (class in stable_baselines.ddpg), 97 load_running_average() (sta-
LnMlpPolicy (class in stable_baselines.deepq), 113 ble_baselines.common.vec_env.VecNormalize
LnMlpPolicy (class in stable_baselines.sac), 148 method), 30
LnMlpPolicy (class in stable_baselines.td3), 162 log_info() (stable_baselines.gail.ExpertDataset
load() (stable_baselines.a2c.A2C class method), 75 method), 52
load() (stable_baselines.acer.ACER class method), 81 logp() (stable_baselines.common.distributions.ProbabilityDistribution
load() (stable_baselines.acktr.ACKTR class method), method), 179
87 LstmPolicy (class in sta-
load() (stable_baselines.common.base_class.BaseRLModel ble_baselines.common.policies), 68
class method), 62
load() (stable_baselines.common.vec_env.VecNormalize M
static method), 30 main() (in module stable_baselines.results_plotter),
load() (stable_baselines.ddpg.DDPG class method), 213
94 make_actor() (stable_baselines.ddpg.CnnPolicy
load() (stable_baselines.deepq.DQN class method), method), 99
110 make_actor() (stable_baselines.ddpg.LnCnnPolicy
load() (stable_baselines.gail.GAIL class method), 122 method), 101
load() (stable_baselines.her.HER class method), 126 make_actor() (stable_baselines.ddpg.LnMlpPolicy
load() (stable_baselines.ppo1.PPO1 class method), method), 98
132 make_actor() (stable_baselines.ddpg.MlpPolicy
load() (stable_baselines.ppo2.PPO2 class method), method), 96
138 make_actor() (stable_baselines.sac.CnnPolicy
load() (stable_baselines.sac.SAC class method), 144 method), 150
load() (stable_baselines.td3.TD3 class method), 158 make_actor() (stable_baselines.sac.LnCnnPolicy
load() (stable_baselines.trpo_mpi.TRPO class method), 151
method), 171 make_actor() (stable_baselines.sac.LnMlpPolicy
load_parameters() (stable_baselines.a2c.A2C method), 148
method), 75 make_actor() (stable_baselines.sac.MlpPolicy
load_parameters() (stable_baselines.acer.ACER method), 147
method), 81 make_actor() (stable_baselines.td3.CnnPolicy
load_parameters() (stable_baselines.acktr.ACKTR method), 163
method), 87 make_actor() (stable_baselines.td3.LnCnnPolicy
load_parameters() (sta- method), 165
ble_baselines.common.base_class.BaseRLModel make_actor() (stable_baselines.td3.LnMlpPolicy
load_parameters() (stable_baselines.ddpg.DDPG make_actor() (stable_baselines.td3.MlpPolicy
load_parameters() (stable_baselines.deepq.DQN make_atari_env() (in module sta-
method), 110 ble_baselines.common.cmd_util), 185
load_parameters() (stable_baselines.gail.GAIL make_critic() (stable_baselines.ddpg.CnnPolicy
load_parameters() (stable_baselines.ppo1.PPO1
228 Index
make_critic() (stable_baselines.ddpg.LnCnnPolicy Monitor (class in stable_baselines.bench.monitor), 190

method), 101 mse() (in module stable_baselines.common.tf_util), 183
make_critic() (stable_baselines.ddpg.LnMlpPolicy mujoco_arg_parser() (in module sta-
method), 98 ble_baselines.common.cmd_util), 186
make_critic() (stable_baselines.ddpg.MlpPolicy MultiCategoricalProbabilityDistribution
method), 96 (class in sta-
make_critics() (stable_baselines.sac.CnnPolicy ble_baselines.common.distributions), 177
method), 150 MultiCategoricalProbabilityDistributionType
make_critics() (stable_baselines.sac.LnCnnPolicy (class in sta-
method), 151 ble_baselines.common.distributions), 178
make_critics() (stable_baselines.sac.LnMlpPolicy
method), 148 N
make_critics() (stable_baselines.sac.MlpPolicy neglogp (stable_baselines.common.policies.ActorCriticPolicy
method), 147 attribute), 66
make_critics() (stable_baselines.td3.CnnPolicy neglogp() (stable_baselines.common.distributions.BernoulliProbabilityD
make_critics() (stable_baselines.td3.LnCnnPolicy neglogp() (stable_baselines.common.distributions.CategoricalProbabili
make_critics() (stable_baselines.td3.LnMlpPolicy neglogp() (stable_baselines.common.distributions.DiagGaussianProbab
make_critics() (stable_baselines.td3.MlpPolicy neglogp() (stable_baselines.common.distributions.MultiCategoricalProb
make_mujoco_env() (in module sta- neglogp() (stable_baselines.common.distributions.ProbabilityDistributio
ble_baselines.common.cmd_util), 185 method), 179
make_proba_dist_type() (in module sta- NormalActionNoise (class in sta-
ble_baselines.common.distributions), 180 ble_baselines.ddpg), 103
make_robotics_env() (in module sta- normalize_obs() (sta-
ble_baselines.common.cmd_util), 185 ble_baselines.common.vec_env.VecNormalize
make_session() (in module sta- method), 31
ble_baselines.common.tf_util), 183 normalize_reward() (sta-
make_vec_env() (in module sta- ble_baselines.common.vec_env.VecNormalize
ble_baselines.common.cmd_util), 186 method), 31
middle_drop() (in module sta- numel() (in module stable_baselines.common.tf_util),
ble_baselines.common.schedules), 189 183
MlpLnLstmPolicy (class in sta-
ble_baselines.common.policies), 70 O
MlpLstmPolicy (class in sta- obs_ph (stable_baselines.common.policies.BasePolicy
ble_baselines.common.policies), 70 attribute), 65
MlpPolicy (class in sta- obs_ph (stable_baselines.ddpg.CnnPolicy attribute),
ble_baselines.common.policies), 69 100
MlpPolicy (class in stable_baselines.ddpg), 96 obs_ph (stable_baselines.ddpg.LnCnnPolicy attribute),
MlpPolicy (class in stable_baselines.deepq), 112 102
MlpPolicy (class in stable_baselines.sac), 146 obs_ph (stable_baselines.ddpg.LnMlpPolicy attribute),
MlpPolicy (class in stable_baselines.td3), 160 98
mode() (stable_baselines.common.distributions.BernoulliProbabilityDistribution
obs_ph (stable_baselines.ddpg.MlpPolicy attribute), 97
method), 174 obs_ph (stable_baselines.deepq.CnnPolicy attribute),
mode() (stable_baselines.common.distributions.CategoricalProbabilityDistribution
115
method), 175 obs_ph (stable_baselines.deepq.LnCnnPolicy at-
mode() (stable_baselines.common.distributions.DiagGaussianProbabilityDistribution
tribute), 116
method), 176 obs_ph (stable_baselines.deepq.LnMlpPolicy attribute),
mode() (stable_baselines.common.distributions.MultiCategoricalProbabilityDistribution
113
method), 178 obs_ph (stable_baselines.deepq.MlpPolicy attribute),
mode() (stable_baselines.common.distributions.ProbabilityDistribution
112
method), 179 obs_ph (stable_baselines.sac.CnnPolicy attribute), 150
Index 229
obs_ph (stable_baselines.sac.LnCnnPolicy attribute), predict() (stable_baselines.acer.ACER method), 82

152 predict() (stable_baselines.acktr.ACKTR method), 87
obs_ph (stable_baselines.sac.LnMlpPolicy attribute), predict() (stable_baselines.common.base_class.BaseRLModel
149 method), 63
obs_ph (stable_baselines.sac.MlpPolicy attribute), 147 predict() (stable_baselines.ddpg.DDPG method), 95
obs_ph (stable_baselines.td3.CnnPolicy attribute), 164 predict() (stable_baselines.deepq.DQN method), 111
obs_ph (stable_baselines.td3.LnCnnPolicy attribute), predict() (stable_baselines.gail.GAIL method), 123
165 predict() (stable_baselines.her.HER method), 127
obs_ph (stable_baselines.td3.LnMlpPolicy attribute), predict() (stable_baselines.ppo1.PPO1 method), 133
163 predict() (stable_baselines.ppo2.PPO2 method), 139
obs_ph (stable_baselines.td3.MlpPolicy attribute), 161 predict() (stable_baselines.sac.SAC method), 145
on_step() (stable_baselines.common.callbacks.BaseCallbackpredict() (stable_baselines.td3.TD3 method), 159
method), 43 predict() (stable_baselines.trpo_mpi.TRPO method),
OrnsteinUhlenbeckActionNoise (class in sta- 171
ble_baselines.ddpg), 103 pretrain() (stable_baselines.a2c.A2C method), 76
outer_scope_getter() (in module sta- pretrain() (stable_baselines.acer.ACER method), 82
ble_baselines.common.tf_util), 183 pretrain() (stable_baselines.acktr.ACKTR method),
88
P pretrain() (stable_baselines.common.base_class.BaseRLModel
param_placeholder() (sta- method), 63
pretrain() (stable_baselines.ddpg.DDPG method),
ble_baselines.common.distributions.ProbabilityDistributionType
method), 179 95
param_shape() (sta- pretrain() (stable_baselines.deepq.DQN method),
111
ble_baselines.common.distributions.BernoulliProbabilityDistributionType
method), 174 pretrain() (stable_baselines.gail.GAIL method), 123
param_shape() (sta- pretrain() (stable_baselines.ppo1.PPO1 method),
133
ble_baselines.common.distributions.CategoricalProbabilityDistributionType
method), 176 pretrain() (stable_baselines.ppo2.PPO2 method),
param_shape() (sta- 139
pretrain() (stable_baselines.sac.SAC method), 145
ble_baselines.common.distributions.DiagGaussianProbabilityDistributionType
method), 177 pretrain() (stable_baselines.td3.TD3 method), 159
param_shape() (sta- pretrain() (stable_baselines.trpo_mpi.TRPO
method), 172
ble_baselines.common.distributions.MultiCategoricalProbabilityDistributionType
method), 178 proba_distribution (sta-
param_shape() (sta- ble_baselines.common.policies.ActorCriticPolicy
attribute), 66
method), 180 proba_distribution_from_flat() (sta-
pdtype (stable_baselines.common.policies.ActorCriticPolicy ble_baselines.common.distributions.DiagGaussianProbabilityDis
PiecewiseSchedule (class in sta- proba_distribution_from_flat() (sta-
ble_baselines.common.schedules), 187 ble_baselines.common.distributions.MultiCategoricalProbabilityD
plot() (stable_baselines.gail.ExpertDataset method), method), 178
52 proba_distribution_from_flat() (sta-
plot_curves() (in module sta- ble_baselines.common.distributions.ProbabilityDistributionType
ble_baselines.results_plotter), 213 method), 180
plot_results() (in module sta- proba_distribution_from_latent() (sta-
ble_baselines.results_plotter), 213 ble_baselines.common.distributions.BernoulliProbabilityDistribu
policy (stable_baselines.common.policies.ActorCriticPolicy method), 174
attribute), 66 proba_distribution_from_latent() (sta-
policy_proba (stable_baselines.common.policies.ActorCriticPolicy ble_baselines.common.distributions.CategoricalProbabilityDistrib
PPO1 (class in stable_baselines.ppo1), 130 proba_distribution_from_latent() (sta-
PPO2 (class in stable_baselines.ppo2), 136 ble_baselines.common.distributions.DiagGaussianProbabilityDis
predict() (stable_baselines.a2c.A2C method), 75 method), 177
230 Index
proba_distribution_from_latent() (sta- ble_baselines.common.distributions.MultiCategoricalProbabilityD

method), 178
ble_baselines.common.distributions.MultiCategoricalProbabilityDistributionType
method), 178 probability_distribution_class() (sta-
proba_distribution_from_latent() (sta- ble_baselines.common.distributions.ProbabilityDistributionType
method), 180
method), 180 ProbabilityDistribution (class in sta-
proba_step() (stable_baselines.common.policies.BasePolicy ble_baselines.common.distributions), 179
method), 65 ProbabilityDistributionType (class in sta-
proba_step() (stable_baselines.common.policies.FeedForwardPolicy ble_baselines.common.distributions), 179
method), 67 processed_obs (sta-
proba_step() (stable_baselines.common.policies.LstmPolicy ble_baselines.common.policies.BasePolicy
proba_step() (stable_baselines.ddpg.CnnPolicy processed_obs (stable_baselines.ddpg.CnnPolicy at-
method), 100 tribute), 100
proba_step() (stable_baselines.ddpg.LnCnnPolicy processed_obs (stable_baselines.ddpg.LnCnnPolicy
proba_step() (stable_baselines.ddpg.LnMlpPolicy processed_obs (stable_baselines.ddpg.LnMlpPolicy
proba_step() (stable_baselines.ddpg.MlpPolicy processed_obs (stable_baselines.ddpg.MlpPolicy at-
method), 97 tribute), 97
proba_step() (stable_baselines.deepq.CnnPolicy processed_obs (stable_baselines.deepq.CnnPolicy
proba_step() (stable_baselines.deepq.LnCnnPolicy processed_obs (sta-
method), 116 ble_baselines.deepq.LnCnnPolicy attribute),
proba_step() (stable_baselines.deepq.LnMlpPolicy 116
method), 113 processed_obs (stable_baselines.deepq.LnMlpPolicy
proba_step() (stable_baselines.deepq.MlpPolicy attribute), 114
method), 112 processed_obs (stable_baselines.deepq.MlpPolicy
proba_step() (stable_baselines.sac.CnnPolicy attribute), 113
method), 150 processed_obs (stable_baselines.sac.CnnPolicy at-
proba_step() (stable_baselines.sac.LnCnnPolicy tribute), 150
method), 152 processed_obs (stable_baselines.sac.LnCnnPolicy
proba_step() (stable_baselines.sac.LnMlpPolicy attribute), 152
method), 149 processed_obs (stable_baselines.sac.LnMlpPolicy
proba_step() (stable_baselines.sac.MlpPolicy attribute), 149
method), 147 processed_obs (stable_baselines.sac.MlpPolicy at-
proba_step() (stable_baselines.td3.CnnPolicy tribute), 147
method), 164 processed_obs (stable_baselines.td3.CnnPolicy at-
proba_step() (stable_baselines.td3.LnCnnPolicy tribute), 164
method), 165 processed_obs (stable_baselines.td3.LnCnnPolicy
proba_step() (stable_baselines.td3.LnMlpPolicy attribute), 166
method), 163 processed_obs (stable_baselines.td3.LnMlpPolicy
proba_step() (stable_baselines.td3.MlpPolicy attribute), 163
method), 161 processed_obs (stable_baselines.td3.MlpPolicy at-
probability_distribution_class() (sta- tribute), 161
method), 175 Q
probability_distribution_class() (sta- q_explained_variance() (in module sta-
ble_baselines.common.distributions.CategoricalProbabilityDistributionType
method), 176
probability_distribution_class() (sta- R
ble_baselines.common.distributions.DiagGaussianProbabilityDistributionType
render() (stable_baselines.common.vec_env.DummyVecEnv
probability_distribution_class() (sta-
Index 231
render() (stable_baselines.common.vec_env.VecEnv method), 176

method), 26 sample_dtype() (sta-
replay_buffer_add() (sta- ble_baselines.common.distributions.DiagGaussianProbabilityDis
ble_baselines.ddpg.DDPG method), 95 method), 177
replay_buffer_add() (sta- sample_dtype() (sta-
ble_baselines.deepq.DQN method), 111 ble_baselines.common.distributions.MultiCategoricalProbabilityD
replay_buffer_add() (stable_baselines.sac.SAC method), 179
method), 146 sample_dtype() (sta-
replay_buffer_add() (stable_baselines.td3.TD3 ble_baselines.common.distributions.ProbabilityDistributionType
reset() (stable_baselines.bench.monitor.Monitor sample_placeholder() (sta-
method), 190 ble_baselines.common.distributions.ProbabilityDistributionType
reset() (stable_baselines.common.vec_env.DummyVecEnv method), 180
method), 28 sample_shape() (sta-
reset() (stable_baselines.common.vec_env.SubprocVecEnv ble_baselines.common.distributions.BernoulliProbabilityDistribu
reset() (stable_baselines.common.vec_env.VecCheckNansample_shape() (sta-
method), 32 ble_baselines.common.distributions.CategoricalProbabilityDistrib
reset() (stable_baselines.common.vec_env.VecEnv method), 176
reset() (stable_baselines.common.vec_env.VecFrameStack ble_baselines.common.distributions.DiagGaussianProbabilityDis
reset() (stable_baselines.common.vec_env.VecNormalizesample_shape() (sta-
method), 31 ble_baselines.common.distributions.MultiCategoricalProbabilityD
reset() (stable_baselines.common.vec_env.VecVideoRecorder method), 179
reset() (stable_baselines.ddpg.NormalActionNoise ble_baselines.common.distributions.ProbabilityDistributionType
reset() (stable_baselines.ddpg.OrnsteinUhlenbeckActionNoise
save() (stable_baselines.a2c.A2C method), 76
method), 103 save() (stable_baselines.acer.ACER method), 82
robotics_arg_parser() (in module sta- save() (stable_baselines.acktr.ACKTR method), 88
ble_baselines.common.cmd_util), 186 save() (stable_baselines.common.base_class.BaseRLModel
rolling_window() (in module sta- method), 63
ble_baselines.results_plotter), 213 save() (stable_baselines.ddpg.DDPG method), 95
save() (stable_baselines.deepq.DQN method), 111
S save() (stable_baselines.gail.GAIL method), 124
SAC (class in stable_baselines.sac), 142 save() (stable_baselines.her.HER method), 127
sample() (in module stable_baselines.common.tf_util), save() (stable_baselines.ppo1.PPO1 method), 133
184 save() (stable_baselines.ppo2.PPO2 method), 139
save() (stable_baselines.sac.SAC method), 146
sample() (stable_baselines.common.distributions.BernoulliProbabilityDistribution
method), 174 save() (stable_baselines.td3.TD3 method), 160
save() (stable_baselines.trpo_mpi.TRPO method), 172
sample() (stable_baselines.common.distributions.CategoricalProbabilityDistribution
method), 175 save_running_average() (sta-
ble_baselines.common.vec_env.VecNormalize
sample() (stable_baselines.common.distributions.DiagGaussianProbabilityDistribution
seed() (stable_baselines.common.vec_env.DummyVecEnv
sample() (stable_baselines.common.distributions.MultiCategoricalProbabilityDistribution
seed() (stable_baselines.common.vec_env.SubprocVecEnv
sample() (stable_baselines.common.distributions.ProbabilityDistribution
sample_dtype() (sta- seed() (stable_baselines.common.vec_env.VecEnv
method), 26
method), 175 seq_to_batch() (in module sta-
sample_dtype() (sta- ble_baselines.common.tf_util), 184
sequential_next()
ble_baselines.common.distributions.CategoricalProbabilityDistributionType (sta-
232 Index
ble_baselines.gail.DataLoader method), method), 82

53 setup_model() (stable_baselines.acktr.ACKTR
set_attr() (stable_baselines.common.vec_env.DummyVecEnv method), 88
method), 28 setup_model() (sta-
set_attr() (stable_baselines.common.vec_env.SubprocVecEnv ble_baselines.common.base_class.BaseRLModel
set_attr() (stable_baselines.common.vec_env.VecEnv setup_model() (stable_baselines.ddpg.DDPG
set_env() (stable_baselines.a2c.A2C method), 76 setup_model() (stable_baselines.deepq.DQN
set_env() (stable_baselines.acer.ACER method), 82 method), 112
set_env() (stable_baselines.acktr.ACKTR method), 88 setup_model() (stable_baselines.gail.GAIL method),
set_env() (stable_baselines.common.base_class.BaseRLModel 124
method), 64 setup_model() (stable_baselines.her.HER method),
set_env() (stable_baselines.ddpg.DDPG method), 95 127
set_env() (stable_baselines.deepq.DQN method), 112 setup_model() (stable_baselines.ppo1.PPO1
set_env() (stable_baselines.gail.GAIL method), 124 method), 133
set_env() (stable_baselines.her.HER method), 127 setup_model() (stable_baselines.ppo2.PPO2
set_env() (stable_baselines.ppo1.PPO1 method), 133 method), 140
set_env() (stable_baselines.ppo2.PPO2 method), 139 setup_model() (stable_baselines.sac.SAC method),
set_env() (stable_baselines.sac.SAC method), 146 146
set_env() (stable_baselines.td3.TD3 method), 160 setup_model() (stable_baselines.td3.TD3 method),
set_env() (stable_baselines.trpo_mpi.TRPO method), 160
172 setup_model() (stable_baselines.trpo_mpi.TRPO
set_random_seed() (stable_baselines.a2c.A2C method), 172
method), 76 shape_el() (in module sta-
set_random_seed() (stable_baselines.acer.ACER ble_baselines.common.distributions), 180
method), 82 single_threaded_session() (in module sta-
set_random_seed() (stable_baselines.acktr.ACKTR ble_baselines.common.tf_util), 184
method), 88 stable_baselines.a2c (module), 71
set_random_seed() (sta- stable_baselines.acer (module), 77
ble_baselines.common.base_class.BaseRLModel stable_baselines.acktr (module), 83
method), 64 stable_baselines.bench.monitor (module),
set_random_seed() (stable_baselines.ddpg.DDPG 190
method), 96 stable_baselines.common.base_class (mod-
set_random_seed() (stable_baselines.deepq.DQN ule), 61
method), 112 stable_baselines.common.callbacks (mod-
set_random_seed() (stable_baselines.gail.GAIL ule), 42
method), 124 stable_baselines.common.cmd_util (mod-
set_random_seed() (stable_baselines.ppo1.PPO1 ule), 185
method), 133 stable_baselines.common.distributions
set_random_seed() (stable_baselines.ppo2.PPO2 (module), 174
method), 139 stable_baselines.common.env_checker
set_random_seed() (stable_baselines.sac.SAC (module), 189
method), 146 stable_baselines.common.evaluation (mod-
set_random_seed() (stable_baselines.td3.TD3 ule), 189
method), 160 stable_baselines.common.policies (mod-
set_random_seed() (sta- ule), 64
ble_baselines.trpo_mpi.TRPO method), stable_baselines.common.schedules (mod-
172 ule), 187
set_venv() (stable_baselines.common.vec_env.VecNormalize stable_baselines.common.tf_util (module),
method), 31 181
setup_model() (stable_baselines.a2c.A2C method), stable_baselines.common.vec_env (module),
76 25
setup_model() (stable_baselines.acer.ACER stable_baselines.ddpg (module), 89
Index 233
stable_baselines.deepq (module), 106 method), 29

stable_baselines.gail (module), 118 step_async() (stable_baselines.common.vec_env.VecCheckNan
stable_baselines.her (module), 124 method), 32
stable_baselines.ppo1 (module), 128 step_async() (stable_baselines.common.vec_env.VecEnv
stable_baselines.ppo2 (module), 134 method), 27
stable_baselines.results_plotter (mod- step_wait() (stable_baselines.common.vec_env.DummyVecEnv
ule), 213 method), 28
stable_baselines.sac (module), 140 step_wait() (stable_baselines.common.vec_env.SubprocVecEnv
stable_baselines.td3 (module), 154 method), 29
stable_baselines.trpo_mpi (module), 167 step_wait() (stable_baselines.common.vec_env.VecCheckNan
start_process() (sta- method), 32
ble_baselines.gail.DataLoader method), step_wait() (stable_baselines.common.vec_env.VecEnv
53 method), 27
step() (stable_baselines.bench.monitor.Monitor step_wait() (stable_baselines.common.vec_env.VecFrameStack
step() (stable_baselines.common.policies.ActorCriticPolicy
step_wait() (stable_baselines.common.vec_env.VecNormalize
step() (stable_baselines.common.policies.BasePolicy step_wait() (stable_baselines.common.vec_env.VecVideoRecorder
step() (stable_baselines.common.policies.FeedForwardPolicy
StopTrainingOnRewardThreshold (class in sta-
method), 68 ble_baselines.common.callbacks), 44
step() (stable_baselines.common.policies.LstmPolicy SubprocVecEnv (class in sta-
method), 69 ble_baselines.common.vec_env), 28
step() (stable_baselines.common.vec_env.VecEnv
method), 27 T
step() (stable_baselines.ddpg.CnnPolicy method), 100 TD3 (class in stable_baselines.td3), 156
step() (stable_baselines.ddpg.LnCnnPolicy method), total_episode_reward_logger() (in module
102 stable_baselines.common.tf_util), 184
step() (stable_baselines.ddpg.LnMlpPolicy method), TRPO (class in stable_baselines.trpo_mpi), 169
99 ts2xy() (in module stable_baselines.results_plotter),
step() (stable_baselines.ddpg.MlpPolicy method), 97 213
step() (stable_baselines.deepq.CnnPolicy method),
115 U
step() (stable_baselines.deepq.LnCnnPolicy method), update_locals() (sta-
116 ble_baselines.common.callbacks.BaseCallback
step() (stable_baselines.deepq.LnMlpPolicy method), method), 43
114
step() (stable_baselines.deepq.MlpPolicy method), V
113 value() (stable_baselines.common.policies.ActorCriticPolicy
step() (stable_baselines.sac.CnnPolicy method), 151 method), 66
step() (stable_baselines.sac.LnCnnPolicy method), value() (stable_baselines.common.policies.FeedForwardPolicy
152 method), 68
step() (stable_baselines.sac.LnMlpPolicy method), value() (stable_baselines.common.policies.LstmPolicy
149 method), 69
step() (stable_baselines.sac.MlpPolicy method), 147 value() (stable_baselines.common.schedules.ConstantSchedule
step() (stable_baselines.td3.CnnPolicy method), 164 method), 187
step() (stable_baselines.td3.LnCnnPolicy method), value() (stable_baselines.common.schedules.LinearSchedule
166 method), 187
step() (stable_baselines.td3.LnMlpPolicy method), value() (stable_baselines.common.schedules.PiecewiseSchedule
163 method), 188
step() (stable_baselines.td3.MlpPolicy method), 161 value() (stable_baselines.ddpg.CnnPolicy method),
step_async() (stable_baselines.common.vec_env.DummyVecEnv 100
method), 28 value() (stable_baselines.ddpg.LnCnnPolicy method),
step_async() (stable_baselines.common.vec_env.SubprocVecEnv 102
234 Index
value() (stable_baselines.ddpg.LnMlpPolicy method),

99
value() (stable_baselines.ddpg.MlpPolicy method), 97
value_flat (stable_baselines.common.policies.ActorCriticPolicy
attribute), 67
value_fn (stable_baselines.common.policies.ActorCriticPolicy
attribute), 67
var_shape() (in module sta-
VecCheckNan (class in sta-
ble_baselines.common.vec_env), 32
VecEnv (class in stable_baselines.common.vec_env), 25
VecFrameStack (class in sta-
VecNormalize (class in sta-
VecVideoRecorder (class in sta-
W
window_func() (in module sta-
ble_baselines.results_plotter), 213
Index 235

Stable Baselines

Uploaded by

Copyright:

Available Formats

Stable Baselines

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stable Baselines

Uploaded by

Copyright:

Available Formats

What is the main purpose of the documentation?

What is the main purpose of the documentation?

What algorithms are covered in the documentation?

What algorithms are covered in the documentation?

Stable Baselines Documentation

Stable Baselines Contributors

Aug 25, 2021

1 Main differences with OpenAI Baselines 3

2 Citing Stable Baselines 215

4 Indices and tables 219

Python Module Index 221

Github repository: https://github.com/hill-a/stable-baselines

Main differences with OpenAI Baselines

brew install cmake openmpi

pip install stable-baselines[mpi]

pip install stable-baselines

1.1.2 Bleeding-edge version

To install the latest master version:

pip install git+https://github.com/hill-a/stable-baselines

1.1.3 Development version

4 Chapter 1. Main differences with OpenAI Baselines

git clone https://github.com/hill-a/stable-baselines && cd stable-baselines

1.1.4 Using Docker Images

Use Built Images

GPU image (requires nvidia-docker):

docker pull stablebaselines/stable-baselines

docker pull stablebaselines/stable-baselines-cpu

Build the Docker Images

Build GPU image (with nvidia-docker):

Build CPU image:

--network=host --build-arg HTTP_PROXY=http://your.proxy.fr:8080/ --build-arg http_

Run the images (CPU/GPU)

Run the nvidia-docker GPU image

˓→baselines bash -c 'cd /root/code/stable-baselines/ && pytest tests/'

Or, with the shell file:

./scripts/run_docker_gpu.sh pytest tests/

Run the docker CPU image

˓→bash -c 'cd /root/code/stable-baselines/ && pytest tests/'

Or, with the shell file:

./scripts/run_docker_cpu.sh pytest tests/

Explanation of the docker command:

1.2 Getting Started

from stable_baselines.common.policies import MlpPolicy

model = PPO2(MlpPolicy, env, verbose=1)

from stable_baselines import PPO2

model = PPO2('MlpPolicy', 'CartPole-v1').learn(10000)

6 Chapter 1. Main differences with OpenAI Baselines

Fig. 1: Define and train a RL agent in one line of code!

1.3 Reinforcement Learning Tips and Tricks

1.3.1 General advice when using Reinforcement Learning

1. Read about RL and Stable Baselines

You have to be aware of the current limitations of reinforcement learning.

1.3. Reinforcement Learning Tips and Tricks 7

How to evaluate an RL algorithm?

1.3.2 Which algorithm should I use?

Note: This covers Discrete, MultiDiscrete, Binary and MultiBinary spaces

8 Chapter 1. Main differences with OpenAI Baselines

Discrete Actions - Single Process