This is a Python wrapper around Naoaki Okazaki (chokkan)'s liblbfgs library of quasi-Newton optimization routines (limited memory BFGS and OWL-QN).

This package aims to provide a cleaner interface to the LBFGS algorithm than is currently available in SciPy, and to provide the OWL-QN algorithm to Python users.

To build PyLBFGS, run python setup.py build_ext.

PyLBFGS is written in Cython and requires setuptools, NumPy, liblbfgs and a relatively recent Cython compiler to build (tested with 0.15.1).

Type:

python setup.py install

(optionally prefixed with sudo) to build and install PyLBFGS.

Type:

python setup.py build_ext -i

to build PyLBFGS in-place, i.e. without installing it.

To run the test suite, make sure you have Nose installed, and type:

nosetests tests/

PyLBFGS was written by Lars Buitinck.

Alexis Mignon submitted a patch for error handling.