RN

3. 4 , C h a
– 3 p te r
.6

Blind S earch

1
 Search Overview
? Introduction to Search

? Blind Search Techniques

aka “Uninformed Search” (Goal vs NonGoal)
? Breadth-First (Uniform Cost)

? Depth-First

? “Iterative Deepening"

? Bi-Directional
? Heuristic Search Techniques
? Stochastic Algorithms
? Game Playing search
? Constraint Satisfaction Problems

2
 Generic Search Algorithm
Searchinsert( start, operations, isGoal ): path
L = make-queue( start )
loop
n := pop( L )
if [ isGoal( n )]
return( n )
S := successors( n, operators )
L := insert( S, L )
until L is empty
return( failure )

insert could be queue, stack, . . .

defines strategy! 3
 Blind S earch

? Blind Search
? Depth-first search
? Breadth-first search
? Iterative deepening
? …
? Not “guided” by goal
? No matter where the goal is,
these algorithms will do the
same thing.
4
 Performance Measures of
S earch Algorithms
? Completeness
Does algorithm always find a sol’n (if ∃)?
? Optimality
Does it always find least cost sol’n?
? Time complexity
How long does it take to find sol’n?
? Space complexity
How much memory is required to find a sol’n?

5
 Parameters b
d

To measure Time and Space complexity:

? b: maximum branching factor of ⋮

the search tree

? Max number of operations at any state
? d: depth of the least-cost solution
? depth of shallowest goal node in search tree
? m: maximum depth of the state space
(may be ∞)
6
Breadth-first search
? Expand shallowest unexpanded node
? Implementation:
? fringe is a FIFO queue,
… new successors go at end

7
Breadth-First Search

8
Properties of Breadth-First search
? Complete? Yes (if b is finite)
? Optimal? Yes (if cost = 1 per step)
? Time? O(bd)
? 1 + b + b2 + … + bd/2 = O(bd)

? Space? O(bd)
? keeps every intermediate node in memory

? Space is the bigger problem (more than time)

9
 Time and Memory R equirements
d #Nodes Time Memory
2 111 .01 msec 11 Kbytes
4 11,111 1 msec 1 Mbyte
6 ~106 1 sec 100 Mb
8 ~108 100 sec 10 Gbytes
10 ~1010 2.8 hours 1 Tbyte
12 ~1012 11.6 days 100 Tbytes
14 ~1014 3.2 years 10,000 Tb
Assumptions: b = 10; 1,000,000 nodes/sec; 100bytes/node
11
Uniform Cost S earch
? BreadthFirst returns SHALLOW-est Goal
… not necessarily best. . .
? Uniform Cost Search:
Expand LEAST Cost node
? If COST ≡ Depth, then UC = BF
? Complete

? Optimal
* *: If g( Successor(n) ) ≥ g(n)
Eg, if g(n) = ∑i c(ai)
? Time: O(b )
d is SUM of arc-costs

? Space: O(b )
d

12
 To insure optimality

? To guarantee OPTIMAL path,

need to maintain queue,
sorted in increasing order:

A S
0
1 10
S 5 B 5 G
A B C
1 5 15
5
15 C
G G
11 10
13
 Uniform-cost search
? Expand least-cost unexpanded node
? Implementation:
? fringe = queue ordered by path cost
? Equivalent to breadth-first if step costs all equal
? Complete? Yes, if step cost ≥ ε
(in trouble if cost = 0)
? Optimal? Yes
…as nodes expanded in increasing order of cost
? Time? O(b⌈C*/ ε⌉)
where C* = cost of optimal solution
? Space? O(b⌈C*/ ε⌉)
15
Depth-first search
? Expand deepest unexpanded node
? Implementation:
? fringe = LIFO queue, i.e., put successors at front

16
Depth-first Search

17
 Properties of Depth-First Search
? Complete? No: fails in infinite-depth spaces, or if loops
? Modify to avoid repeated states along path
? complete in finite spaces

? Optimal? No
… first found ≠? best 20
⋮

? Time? O( bm) :
?terrible if m is much larger than d
?but if solutions are dense, may be much faster than BF

? Space? ( b m)
? d=12 ⇒ 12 kb, not 111 terabytes! 18
 BFS/UC vs. DFS
Complete? Optimal? Time Space
BFS/UC YES YES bd bd
DFS finite depth NO bm bm

? Time:
? m=d DFS typically wins
:might winguara n t e e s ,
g
m>dllenBFS
a e s
Cmh= ’
?

e t B F S o r y ? ?
?
∞
w to g
BFS probably
s m em
better
Ho ’
?

n l y D F S
? Space
u si n g o
? DFS almost always beats BFS
19
Which Strategy to Use?
? Depends on problem.
? If there are infinite paths
⇒ depth-first is bad
? If goal is at known depth
⇒ depth-first is good
? If ∃ large (possibly ∞) branching factor
⇒ breadth-first is bad

(Could try nondeterministic search:

Expand an open node at random.)

20
Depth- Limited S trategy
? Depth-first with depth cut-off k
(do NOT expand nodes below depth k)

? Three possible outcomes:

? Solution
? Failure (no solution)
? Cutoff (no solution within cutoff)

21
 Depth-Limited Depth-First-Search
? Depth cut-off: k=3

? Complete: No unless soln @ depth ≤ k

? Optimal: No
? Time: O(bk)
? Space: O(b k)
22
 Iterative Deepening S trategy

Use an artificial depth cutoff, k.

? For k = 1…
? Use Depth-limited Depth-First Search(k)
? If succeeds: DONE.

? If not: increase k by 1
(Regenerate nodes, as necessary)

23
Iterative Deepening Search k=0

24
Iterative Deepening Search: k=1

25
Iterative Deepening Search: k =2

26
Iterative Deepening Search k=3

27
28
 Iterative Deepening: Analysis
? Time: ≈ BFS !
? … even though it regenerates intermediate nodes !
? Why? Almost all work ON FINAL LEVEL anyway!
? Eg: b = 10, d = 5:
? BFS expands 1 + 10 + 100 + … + 100,000 = 111,111
? IDS expands
? bottom level: 1 time
? second to bottom: 2 times
? …
? toplevel: d+1 times
total:(d+1)b0 + d b1 + (d-1)b2 + … + 3bd-2 +2bd-1 + 1bd
… 100,000 + 20,000 + …+ 50 + 6 = 123,456
? Ratio of IDS to BFS: ≈ [b / (b-1)]2
? Cost of repeating work at shallow depths: MINOR!
29
 Properties of
Iterative Deepening S earch

? Complete? Yes
? Optimal? Yes, if step cost = 1
? Time?
(d+1)b0 + d b1 + (d-1)b2 + … + bd = O(bd)
? Space? O(b d)
? IDS does not get stuck on infinite path
? Space: Same as DFS – but with d, not m
(as each search is DFS) 32
 BiDirectional
S earch
? Simultaneously:
? Search “forward" from start
? Search “backward" from goal
Stop when two searches meet in middle
? If branching factor = b in each direction & solution at depth d
⇒ need only O(2bd/2) = O(bd/2) steps

? Eg: b = 10, d = 6:
BFS expands 1,111,111 nodes
BiDirectional: 2,222 !
? Issues:
? How to “search backwards from goal"?
? What if > 1 goals (chess)?
? How to check if paths meet? constant time?
? What type of search done in each half? (BFS)

33
34
Comparing “Blind" Search
Strategies

35
Comparison of S trategies
? Breadth-first is complete and optimal,
but has high space complexity
? Bad when branching factor is high
? Depth-first is space efficient,
but not complete nor optimal
? Bad when search depth is infinite
? Iterative deepening is asymptotically
optimal !

36
 Avoiding Repeated States
? May reach same state thru multiple paths

? ... if operations are REVERSIBLE (∞)

? ⇒ “Obvious" algorithms may
? be inefficient (exponentially worse)
? loop forever!
37
 R epeated S tates
No Few Many

1 2 3
search tree is finite search tree is infinite
4 5
7 8 6

8-queens assembly 8-puzzle and robot navigation

planning
38
 Approaches to Deal
w/Repeated State
? Don't return to parent state
? Don't generate successor ≡ node's parent
? Don't allow cycles
? Don't generate successor ≡ node's ancestor
? Don't ever revisit state
? Keep every visited state in memory! O(bd)

39
S ummary of Blind S earch
? Search strategies:
? breadth-first, depth-first, iterative deepening, …
? Evaluation of strategies:
? completeness, optimality, time and space
complexity
? Iterative deepening search
? uses only linear space
? ≈ same time as other blind-searchers
? Avoid repeated states

40
41