Mychapter 3 A

Informed search algorithms
CHAPTER 4
Oliver Schulte
Summer 2011
Outline
Best-first search
A* search
Heuristics
Local search algorithms
Hill-climbing search
Simulated annealing search
Local beam search
Environment Type Discussed In this Lecture
3
Fully
Observable Static Environment
yes
Deterministic
yes
Sequential
yes no
Discrete no
yes Discrete
no
yes
Planning, Control, Vector Search: Continuous Function
heuristic cybernetics Constraint Optimization
search Satisfaction
CMPT 310 - Blind Search
Review: Tree search
A search strategy is defined by picking the order of

node expansion
Which nodes to check first?
Knowledge and Heuristics
Simon and Newell, Human Problem Solving, 1972.

Thinking out loud: experts have strong opinions like
“this looks promising”, “no way this is going to
work”.
S&N: intelligence comes from heuristics that help
find promising states fast.
Best-first search
 Idea: use an evaluation function f(n) for each node

 estimate of "desirability"
 Expand most desirable unexpanded node
 Implementation:
Order the nodes in frontier in decreasing order of
desirability
 Special cases:
 greedy best-first search
 A* search



Romania with step costs in km
Greedy best-first search
Evaluation function
 f(n) = h(n) (heuristic)
 = estimate of cost from n to goal
e.g., hSLD(n) = straight-line distance from n to

Bucharest
Greedy best-first search expands the node that
appears to be closest to goal

Greedy best-first search example
http://aispace.org/search/
Properties of greedy best-first search
Complete? No – can get stuck in loops,

 e.g. as Oradea as goal
 Iasi  Neamt  Iasi  Neamt 
Time? O(bm), but a good heuristic can give dramatic
improvement
Space? O(bm) -- keeps all nodes in memory
Optimal? No

A* search
Idea: avoid expanding paths that are already

expensive.
Very important!
Evaluation function f(n) = g(n) + h(n)

h(n) = estimated cost from n to goal
f(n) = estimated total cost of path through n
to goal
g(n) = cost so far to reach n
A* search example
A* search example
A* search example
A* search example
A* search example
A* search example
http://aispace.org/search/
• We stop when the node with the lowest f-value

is a goal state.
• Is this guaranteed to find the shortest path?
Admissible heuristics
A heuristic h(n) is admissible if for every node n,

h(n) ≤ h*(n), where h*(n) is the true cost to reach the
goal state from n.
An admissible heuristic never overestimates the cost
to reach the goal, i.e., it is optimistic.
Example: h (n) (never overestimates the actual

SLD
road distance)
Negative Example: Fly heuristic: if wall is dark, then
distance from exit is large.
Theorem: If h(n) is admissible, A* using TREE-
SEARCH is optimal

Optimality of A* (proof)
 Suppose some suboptimal goal path G has been generated and is in the
2
frontier. Let n be an unexpanded node in the frontier such that n is on a
shortest path to an optimal goal G.
 f(G ) = g(G ) since h(G2) = 0 because h is admissible

2 2
 g(G ) > g(G) since G2 is suboptimal, cost of reaching G is
2
less.
 f(G) = g(G) since h(G) = 0
 f(G ) > f(G) from above
2

Optimality of A* (proof)
 Suppose some suboptimal goal path G has been generated and is in

2
the frontier. Let n be an unexpanded node in the frontier such that n
is on a shortest path to an optimal goal G.
 f(G ) > f(G) from above

2
 h(n) ≤ h*(n) since h is admissible, h* is minimal distance.
 g(n) + h(n) ≤ g(n) + h*(n)
 f(n) ≤ f(G)
Hence f(G2) > f(n), and A* will never select G2 for expansion


Consistent heuristics
 A heuristic is consistent if for every node n, every successor n' of n generated by
any action a,
h(n) ≤ c(n,a,n') + h(n')

 Intuition: can’t do worse than going through n’.
 If h is consistent, we have
f(n') = g(n') + h(n') = g(n) + c(n,a,n') + h(n')
≥ g(n) + h(n) = f(n)
 i.e., f(n) is non-decreasing along any path.
 Theorem: If h(n) is consistent, A* using GRAPH-SEARCH is optimal




Optimality of A*
 A* expands nodes in order of increasing f value

 http://aispace.org/search/
 Gradually adds "f-contours" of nodes
 Contour i has all nodes with f=fi, where fi < fi+1


Properties of A*
Complete? Yes (unless there are infinitely many

nodes with f ≤ f(G) )
Time? Exponential
Space? Keeps all nodes in memory
Optimal? Yes




E.g., for the 8-puzzle:

 h (n) = number of misplaced tiles
1
 h (n) = total Manhattan distance
2
(i.e., no. of squares from desired location of each tile)
h (S) = ?
1
h (S) = ?
2



E.g., for the 8-puzzle:

 h (n) = number of misplaced tiles
1
 h (n) = total Manhattan distance
2
(i.e., no. of squares from desired location of each tile)
h (S) = ? 8
1
h (S) = ? 3+1+2+2+2+3+3+2 = 18
2


Dominance
 If h (n) ≥ h (n) for all n (both admissible) then h dominates h .

2 1 2 1
 h is better for search
2
 Typical search costs (average number of nodes expanded):
 d=12 IDS = 3,644,035 nodes

A*(h1) = 227 nodes
A*(h2) = 73 nodes
 d=24 IDS = too many nodes
A*(h1) = 39,135 nodes
A*(h2) = 1,641 nodes



Relaxed problems
A problem with fewer restrictions on the actions is

called a relaxed problem
The cost of an optimal solution to a relaxed problem is
an admissible heuristic for the original problem
If the rules of the 8-puzzle are relaxed so that a tile can
move anywhere, then h1(n) gives the shortest solution
If the rules are relaxed so that a tile can move to any
adjacent square, then h2(n) gives the shortest solution




Summary
Heuristic functions estimate costs of shortest paths

Good heuristics can dramatically reduce search cost
Greedy best-first search expands lowest h
 incomplete and not always optimal
A∗ search expands lowest g + h
 complete and optimal
 also optimally efficient (up to tie-breaks)
Admissible heuristics can be derived from exact
solution of relaxed problems
Missionaries and Cannibals
Old puzzle: has been around since 700 AD.

Solved by Computer!
Try it at home!
Good for depth-first search: basically, linear solution
path.
Another view of informed search: we use so much
domain knowledge and constraints that depth-first
search suffices.
The problem graph is larger than the problem statement.
 Taking the state graph as input seems problematic.

Mychapter 3 A

Uploaded by

Copyright:

Available Formats

Mychapter 3 A

Uploaded by

Document Information

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

Mychapter 3 A

Uploaded by

Copyright:

Available Formats

Informed search algorithms

A search strategy is defined by picking the order of

Simon and Newell, Human Problem Solving, 1972.

 Idea: use an evaluation function f(n) for each node

e.g., hSLD(n) = straight-line distance from n to

Complete? No – can get stuck in loops,

Idea: avoid expanding paths that are already

Evaluation function f(n) = g(n) + h(n)

• We stop when the node with the lowest f-value

A heuristic h(n) is admissible if for every node n,

Example: h (n) (never overestimates the actual

 f(G ) = g(G ) since h(G2) = 0 because h is admissible

 Suppose some suboptimal goal path G has been generated and is in

 f(G ) > f(G) from above

h(n) ≤ c(n,a,n') + h(n')

 A* expands nodes in order of increasing f value

Complete? Yes (unless there are infinitely many

E.g., for the 8-puzzle:

E.g., for the 8-puzzle:

 If h (n) ≥ h (n) for all n (both admissible) then h dominates h .

 Typical search costs (average number of nodes expanded):

 d=12 IDS = 3,644,035 nodes

A problem with fewer restrictions on the actions is

Heuristic functions estimate costs of shortest paths

Old puzzle: has been around since 700 AD.

You might also like