Outline

CS380: Intro to Artificial

Problem Solving in Deterministic, Single-
Intelligence Agent Domains
9/29/2006  Uninformed Search Methods
 BFS, DFS, IDS
Jay Modi
Department of Computer Science
Drexel University
www.cs.drexel.edu/~pmodi

An Example Search Problem

Terminology
 b: branching factor, max number of next
states of any state
 d: steps to goal
 m: maximum possible path length
 Not always finite. Why not?

1
 Terminology: States vs. Nodes
Tree search algorithms
 A state is a property of the world, e.g., location
 A node is a data structure which specifies state, parent node, action,  Basic idea:
path cost g(x), depth  offline, simulated exploration of state space by
 A state can have multiple nodes
generating successors of already-explored states
Nodes (a.k.a.~expanding states)
State State: Sibiu
Sibiu State of Parent Node: Arad

State: Sibiu
Arad Fagaras State of Parent Node: Fagaras

The Mother of All Search

Complications
Algorithms
 current state <− start state  How to choose next state from fringe set?
 loop:  Answer has significant implications on:
 is current state = the goal state? If yes, output path and  Time complexity
 Number of loops
terminate.
 Space complexity
 generate all next states of current state and add to  Size of fringe set
fringe set
 Completeness
 current state <− choose state from fringe set  Does loop terminate
 Optimality
All search algorithms can be viewed as instances of this.  Quality of output path
Simple, right?

2
How to Choose From Fringe Set? Breadth-First Search
 One idea:
 Implement fringe set as a First-in First-out
(FIFO) queue
f -> e,d,c,b -> a
 We call this Breadth-First Search

Start State: A
Current State: A
Fringe Set: {}

Breadth-First Search Breadth-First Search

Start State: A Start State: A

Current State: A Current State: B
Fringe Set: {C, B} Fringe Set: { E, D, C}
Choose next state: B Choose next state: C

3
 Breadth-First Search Properties of BFS
 Is it Complete?
 Is it Optimal?
 Will it always return shortest path?
 Space complexity
 How large can the fringe set get?
 O(bd+1) :( Why?
Start State: A
Current State: C  Time complexity?
Fringe Set: { G, F, E, D}
Choose next state: D

How to Choose From Fringe Set? Depth-First Search

 Another idea:
 Implement fringe set as a Last-in,First-out
(LIFO) queue
f -> e,d,c,b,a

f <- f,e,d,c,b,a

Start State: A
 We call this Depth-First Search Goal State: M
Current State: A
Fringe Set: {}

4
 Depth-First Search Depth-First Search

Start State: A Start State: A

Current State: A Current State: B
Fringe Set: {B, C} Fringe Set: {D, E, C}
Choose Next State: B Choose Next State: D

Depth-First Search Depth-First Search

Start State: A Q: What would Start State: A

Current State: B BFS do? Current State: D
Fringe Set: {D, E, C} Fringe Set: {H,I,E,C}
Choose Next State: D Choose Next State: H

5
 Depth-First Search Depth-First Search

Start State: A Start State: A

Current State: H Current State: I
Fringe Set: {I,E,C} Fringe Set: {E,C}
Choose Next State: I Choose Next State: E

Depth-First Search Depth-First Search

Start State: A Start State: A

Current State: E Current State: J
Fringe Set: {J, K ,C} Fringe Set: {K ,C}
Choose Next State: J Choose Next State: K

6
 Depth-First Search Depth-First Search

Start State: A Start State: A

Current State: K Current State: C
Fringe Set: {C} Fringe Set: {F, G}
Choose Next State: C Choose Next State: F

Depth-First Search Depth-First Search

Start State: A Start State: A

Current State: F Current State: L
Fringe Set: {L, M, G} Fringe Set: {M, G}
Choose Next State: L Choose Next State: M

7
 Depth-First Search Properties of depth-first search

 Space?
 bm+1
 Time?
 Optimal?
 Complete?
 No. Why?
Start State: A
Current State: M Goal Reached.  Can we fix this?
Fringe Set: {G} Algorithm Terminates
Current State = Goal State

Properties of depth-first search Guaranteeing Completeness

 Complete? No: fails in infinite-depth spaces,  Idea: Invoke Depth-Limited DFS with
spaces with loops
iteratively increasing values of L
 Modify to avoid repeated states along path
 complete in finite spaces  We call this Iterative Deepening Search
 Idea: Cut off search after depth limit L
 We call this Depth-Limited DFS
 Is it complete?

8
Iterative deepening search l =0 Iterative deepening search l =1

Iterative deepening search l =2 Iterative deepening search l =3

9
 Iterative deepening search Iterative deepening search
 Number of nodes generated in a depth-limited search to depth d with
 Iterative Deepening search repeatedly re- branching factor b?
 NDLS = b0 + b1 + b2 + … + bd-2 + bd-1 + bd
expanding nodes
 Number of nodes generated in an iterative deepening search to depth d with
 Seems inefficient, but is it? branching factor b?
 NIDS = (d+1)b0 + d b^1 + (d-1)b^2 + … + 3bd-2 +2bd-1 + 1bd

 For b = 10, d = 5,
 NDLS = 1 + 10 + 100 + 1,000 + 10,000 + 100,000 = 111,111
 NIDS = 6 + 50 + 400 + 3,000 + 20,000 + 100,000 = 123,456

 Overhead = (123,456 - 111,111)/111,111 = 11%

 Iterative Deepening is the method of choice for searching large state spaces
when depth of solution is not known

Properties of iterative deepening Properties of iterative deepening

search search
 Complete?  Complete? Yes
 Time?  Time? d b1 + (d-1)b2 + … + bd = O(bd)
 Space?  Space? O(bd)
 Optimal?  Optimal? Yes, if step cost = 1

10
 Bidirectional Search Bidirectional Search
 In IDS, we saw that branching factor gets  In IDS, we saw that branching factor gets
exponentially worse with depth exponentially worse with depth
 Idea: Minimize depth by doing two parallel
searches

Path Costs How to Choose From Fringe Set?

 So far, we have assumed unit cost for each  A third idea:
state transition.  Implement fringe set as an insert-in-order
Priority queue
 What if each transition has a different cost?
(e,6)

(b,10), (a,5), (d,3), (c,1)

 We call this Uniform-Cost Search

11
 Uniform Cost Search Uniform Cost Search

1 3 1 3

3 4 0 4 3 4 0 4

Start State: A Start State: A

Current State: A Current State: A
Fringe Set: {} Fringe Set: {(C,3), (B,1)}
Choose next state: B

Uniform Cost Search Uniform Cost Search

1 3 1 3

3 4 0 4 3 4 0 4

Start State: A Start State: A

Current State: B Current State: C
Fringe Set: { (E,5), (D,4), (C,3)} Fringe Set: { (G, 7), (E,5), (D,4), (F,3)}
Choose next state: C Choose next state: F

12
Summary of algorithms Summary
 Agent and Environment Types

 Variety of uninformed search strategies

 Iterative deepening search uses only linear space and not

much more time than other uninformed algorithms

13