Iterative Deepening and IDA*

Alan Mackworth

UBC CS 322 – Search 6

January 21, 2013

Textbook § 3.7.3
 Lecture Overview

•  Recap from last week

•  Iterative Deepening

Slide 2
 Search with Costs
•  Sometimes there are costs associated with arcs.

Def.: The cost of a path is the sum of the costs of its arcs
k
cost ( n0 ,…, nk ) = ∑ cost( ni −1 , ni )
i =1

•  In this setting we often don't just want to find any solution

–  we usually want to find the solution that minimizes cost

Def.: A search algorithm is optimal if

when it finds a solution, it is the best one:
it has the lowest path cost

3
 Lowest-Cost-First Search (LCFS)
•  Expands the path with the lowest cost on the frontier.

•  The frontier is implemented as a priority queue ordered by

path cost.
•  How does LCFS differ from Dijkstra’s shortest path
algorithm?
-  The two algorithms are very similar
-  But Dijkstra’s algorithm
-  computes shortest distance from one node to all other nodes
-  works with nodes not with paths
-  stores one bit per node (infeasible for infinite/very large graphs)
-  checks for cycles

4
 Heuristic search
Def.:
A search heuristic h(n) is an estimate of the cost of the optimal
(cheapest) path from node n to a goal node.

Estimate: h(n1)
n1

n2 Estimate: h(n2)

n3
Estimate: h(n3)

5
 Best-First Search (LCFS)
•  Expands the path with the lowest h value on the frontier.

•  The frontier is implemented as a priority queue ordered by

h.

•  Greedy: expands path that appears to lead to the goal

quickest
-  Can get trapped
-  Can yield arbitrarily poor solutions
-  But with a perfect heuristic, it moves straight to the goal

6
 A*
•  Expands the path with the lowest cost + h value on the
frontier

•  The frontier is implemented as a priority queue ordered by

f(p) = cost(p) + h(p)

7
 Admissibility of a heuristic
Def.:
Let c(n) denote the cost of the optimal path from node n to any
goal node. A search heuristic h(n) is called
admissible if h(n) ≤ c(n) for all nodes n, i.e. if for all nodes it
is an underestimate of the cost to any goal.

•  E.g. Euclidean distance in routing networks

•  General construction of heuristics: relax the problem,
i.e. ignore some constraints
-  Can only make it easier
-  Saw lots of examples on Wednesday:
Routing network, grid world, 8 puzzle, Infinite Mario

8
 Admissibility of A*
•  A* is complete (finds a solution, if one exists) and
optimal (finds the optimal path to a goal) if:

•  the branching factor is finite

•  arc costs are > ε > 0
•  h is admissible.

•  This property of A* is called admissibility of A*

9
 Why is A* admissible: complete
If there is a solution, A* finds it:
-  fmin:= cost of optimal solution path s (unknown but finite)

-  Lemmas for prefix pr of s (exercise: prove at home)

-  Has cost f(pr) ≤ fmin (due to admissibility)
-  Always one such pr on the frontier (prove by induction)

-  A* only expands paths with f(p) ≤ fmin

-  Expands paths p with minimal f(p)
-  Always a pr on the frontier, with f(pr) ≤ fmin
-  Terminates when expanding s

-  Number of paths p with cost f(p) ≤ fmin is finite

-  Let cmin > 0 be the minimal cost of any arc
-  k := fmin / cmin. All paths with length > k have cost > fmin
-  Only bk paths of length k. Finite b ⇒ finite
 Why is A* admissible: optimal
New Proof (by contradiction)
–  Assume hypothesis (for contradiction):
First solution s’ that A* expands is suboptimal: i.e. cost(s’) > fmin

–  Since s’ is a goal, h(s’) = 0, and f(s’) = cost(s’) > fmin

–  A* selected s’ ⇒ all other paths p on the frontier

had f(p) ≥ f(s’) > fmin

–  But we know that a prefix pr of optimal solution path s is on the

frontier, with f(pr) ≤ fmin
⇒ Contradiction!
–  QED
Summary: any prefix of optimal solution is expanded before suboptimal
solution would be expanded
11
 Learning Goals for last week

•  Select the most appropriate algorithms for specific

problems
–  Depth-First Search vs. Breadth-First Search
vs. Least-Cost-First Search vs. Best-First Search vs. A*
•  Define/read/write/trace/debug different search algorithms
-  With/without cost
-  Informed/Uninformed
•  Construct heuristic functions for specific search problems
•  Formally prove A* completeness and optimality
-  Define optimal efficiency

12
 Learning Goals for last week, continued
•  Apply basic properties of search algorithms:
–  completeness, optimality, time and space complexity
Complete Optimal Time Space
DFS N N O(bm) O(mb)
(Y if finite & no
cycles)
BFS Y Y O(bm) O(bm)
LCFS Y Y  m)
O(b O(bm)
(when arc costs available) Costs > 0 Costs ≥ 0
Best First N N  m)
O(b O(bm)
(when h available)
A* Y Y  m)
O(b O(bm)
(when arc costs and h Costs > 0
 Costs ≥ 0
available) h admissible h admissible
13
 Lecture Overview

•  Recap from last week

•  Iterative Deepening

14
 Iterative Deepening DFS (IDS): Motivation
Want low space complexity but completeness and optimality
Key Idea: re-compute elements of the frontier
rather than saving them
Complete Optimal Time Space
DFS N N O(bm) O(mb)
(Y if finite & no
cycles)
BFS Y Y O(bm) O(bm)
LCFS Y Y  m)
O(b O(bm)
(when arc costs available) Costs > 0 Costs ≥ 0
Best First N N  m)
O(b O(bm)
(when h available)
A* Y Y  m)
O(b O(bm)
(when arc costs and h Costs > 0
 Costs ≥ 0
available) h admissible h admissible
15
 Iterative Deepening DFS (IDS) in a Nutshell
•  Use DFS to look for solutions at depth 1, then 2, then 3, etc
–  For depth D, ignore any paths with longer length
–  Depth-bounded depth-first search

depth = 1

depth = 2

depth = 3

...
 (Time) Complexity of IDS
•  That sounds wasteful!
•  Let’s analyze the time complexity
•  For a solution at depth m with branching factor b

Depth Total # of paths #times created by #times created Total #paths 


at that level BFS (or DFS) by IDS for IDS
1 b 1 m mb
2 b2 1 m-1 (m-1) b2
. . . . .
. . . . .
. . . . .
m-1 bm-1 1 2 2 bm-1
m bm 1 1 bm
17
 (Time) Complexity of IDS
Solution at depth m, branching factor b
Total # of paths generated:
bm + 2 bm-1 + 3 bm-2 + ...+ mb
= bm (1 b0 + 2 b-1 + 3 b-2 + ...+ m b1-m )
m m
= b m (∑ ib1−i ) = b m (∑ i (b −1 )i −1 )
i =1 i =1 2 2
m ⎛ b ⎞
∞
m −1 i −1 ⎛ 1 ⎞
≤ b (∑ i (b ) ) = b ⎜ m
−1 ⎟
= b ⎜ m
⎟ ∈ O(b )
i =0 ⎝ 1 − b ⎠ ⎝ b − 1 ⎠
∞
i 1
Geometric progression: for |r|<1: ∑ r =
i =0 1− r
d ∞
i
∞
i −1 1
∑ r = ∑ ir = 2
dr i =0 i =0 (1 − r )
 Further Analysis of Iterative Deepening DFS (IDS)
•  Space complexity

O(bm) O(mb) O(bm) O(b+m)

–  DFS scheme, only explore one branch at a time

•  Complete? Yes No

–  Only finite # of paths up to depth m, doesn’t explore longer paths

•  Optimal? Yes No

–  Proof by contradiction

19
 Search methods so far
Complete Optimal Time Space
DFS N N O(bm) O(mb)
(Y if finite & no
cycles)
BFS Y Y O(bm) O(bm)
IDS Y Y O(bm) O(mb)
LCFS Y Y  m)
O(b O(bm)
(when arc costs available) Costs > 0 Costs >=0
Best First N N  m)
O(b O(bm)
(when h available)
A* Y Y  m)
O(b O(bm)
(when arc costs and h Costs > 0
 Costs >=0
available) h admissible h admissible

20
(Heuristic) Iterative Deepening: IDA*

•  Like Iterative Deepening DFS

–  But the depth bound is measured in terms of the f value

•  If you don’t find a solution at a given depth

–  Increase the depth bound:
to the minimum of the f-values that exceeded the previous bound

21
Analysis of Iterative Deepening A* (IDA*)
•  Complete and optimal? Same conditions as A*
–  h is admissible
–  all arc costs > ε > 0
–  finite branching factor

 m)
•  Time complexity: O(b

•  Space complexity:

O(bm) O(mb) O(bm) O(b+m)

–  Same argument as for Iterative Deepening DFS

22
 Search methods so far
Complete Optimal Time Space
DFS N N O(bm) O(mb)
(Y if no cycles)
BFS Y Y O(bm) O(bm)
IDS Y Y O(bm) O(mb)
LCFS Y Y  m)
O(b O(bm)
(when arc costs available) Costs > 0 Costs >=0
Best First N N  m)
O(b O(bm)
(when h available)
A* Y Y  m)
O(b O(bm)
(when arc costs and h Costs > 0
 Costs >=0
available) h admissible h admissible
IDA* Y (same cond. Y  m)
O(b O(mb)
as A*)
 Learning Goals for today’s class
•  Define/read/write/trace/debug different search algorithms
-  New: Iterative Deepening,
Iterative Deepening A*
•  Apply basic properties of search algorithms:
–  completeness, optimality, time and space complexity

Announcements:
–  Practice exercises on course home page
•  Heuristic search
•  Please use these! (Only takes 5 min. if you understood things…)
–  Assignment 1 is out: see Connect.

24