Scribd
Upload
22 views41 pages

CSC3402 Lecture4 InformedSearch

Uploaded by

Godfrey Bwalya
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
22 views41 pages

CSC3402 Lecture4 InformedSearch

Uploaded by

Godfrey Bwalya
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 41

CSC3402 – Introduction to AI

Lesson 4: Informed search


algorithms
Material

 Chapter 4 Section 1 - 3
Outline

 Best-first search
 Greedy best-first search
 A* search
 Heuristics
 Local search algorithms
 Hill-climbing search
 Simulated annealing search
 Local beam search
 Genetic algorithms
Review: Tree search
 A search strategy is defined by picking the order of
node expansion
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 fringe 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
Greedy best-first search
example
Greedy best-first search
example
Greedy best-first search
example
Properties of greedy best-
first search
 Complete? No – can get stuck in loops, e.g., 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
 Evaluation function f(n) = g(n) + h(n)
 g(n) = cost so far to reach n
 h(n) = estimated cost from n to goal
 f(n) = estimated total cost of path through n to goal
A* search example
A* search example
A* search example
A* search example
A* search example
A* search example
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: hSLD(n) (never overestimates the
actual road distance)
 Theorem: If h(n) is admissible, A* using
TREE-SEARCH is optimal

Optimality of A* (proof)
 Suppose some suboptimal goal G2 has been
generated and is in the fringe. Let n be an
unexpanded node in the fringe such that n is on a
shortest path to an optimal goal G.

 f(G2) = g(G2) since h(G2) = 0


 g(G2) > g(G) since G2 is suboptimal
 f(G) = g(G) since h(G) = 0
 f(G2) > f(G) from above
Optimality of A* (proof)
 Suppose some suboptimal goal G2 has been generated and is in the fringe. Let n be
an unexpanded node in the fringe such that n is on a shortest path to an optimal goal
G.

 f(G2) > f(G) from above


 h(n) ≤ h^*(n) since h is admissible
 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')

 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



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
Admissible heuristics
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


Admissible heuristics
E.g., for the 8-puzzle:
 h1(n) = number of misplaced tiles
 h2(n) = total Manhattan distance
(i.e., no. of squares from desired location of each tile)

 h1(S) = ? 8
 h2(S) = ? 3+1+2+2+2+3+3+2 = 18


Dominance

 If h (n) ≥ h (n) for all n (both admissible)


2 1
 then h dominates h
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
Local search algorithms

 In many optimization problems, the


path to the goal is irrelevant; the goal
state itself is the solution

 State space = set of "complete"


configurations
 Find configuration satisfying
constraints, e.g., n-queens

 In such cases, we can use local search


algorithms
 keep a single "current" state, try to
improve it
Example: n-queens

 Put n queens on an n × n board with no two queens


on the same row, column, or diagonal
Hill-climbing search

 "Like climbing Everest in thick fog with amnesia"


Hill-climbing search

 Problem: depending on initial state, can get stuck in


local maxima


Hill-climbing search: 8-
queens problem

 h = number of pairs of queens that are attacking each other, either directly or
indirectly
 h = 17 for the above state

Hill-climbing search: 8-
queens problem

• A local minimum with h = 1


•
Simulated annealing search

 Idea: escape local maxima by


allowing some "bad" moves but
gradually decrease their
frequency


Properties of simulated
annealing search
 One can prove: If T decreases
slowly enough, then simulated
annealing search will find a global
optimum with probability
approaching 1

 Widely used in VLSI layout, airline


scheduling, etc
Local beam search

 Keep track of k states rather than just


one

 Start with k randomly generated states

 At each iteration, all the successors of


all k states are generated

 If any one is a goal state, stop; else


select the k best successors from the
complete list and repeat.
Genetic algorithms

 A successor state is generated by combining


two parent states

 Start with k randomly generated states


(population)

 A state is represented as a string over a finite


alphabet (often a string of 0s and 1s)

 Evaluation function (fitness function). Higher


values for better states.

 Produce the next generation of states by


selection, crossover, and mutation
Genetic algorithms

 Fitness function: number of non-attacking


pairs of queens (min = 0, max = 8 × 7/2 =
28)
 24/(24+23+20+11) = 31%
 23/(24+23+20+11) = 29% etc
Genetic algorithms
Summary

 A heuristic is and estimation of a cost from node to


goal state
 Heuristics improve the quality of search
 Informed search algorithms use heuristics
 Greedy Best-First Search uses heuristics entirely
 A* uses heuristics and actual cost
 Local search algorithms do not care about the path to
goal, but the goal itself.
 Hill-climbing
 Simulated Annealing
 Local beam
 GA

You might also like