Solving problems by searching

1
 Why Search?
 To achieve goals or to maximize our utility
we need to predict what the result of our
actions in the future will be.

 There are many sequences of actions, each

with their own utility.

 We want to find, or search for, the best one.

2
 Example: Romania
 On holiday in Romania; currently in Arad.
 Flight leaves tomorrow from Bucharest
 Formulate goal:
 be in Bucharest
 Formulate problem:
 states: various cities
 actions: drive between cities or choose next city
 Find solution:
 sequence of cities, e.g., Arad, Sibiu, Fagaras, Bucharest

3
Example: Romania

4
 Task Environment
 Static / Dynamic
Previous problem was static: no attention to changes in environment
 Observable / Partially Observable / Unobservable
Previous problem was observable: it knew its states at all times.
 Deterministic / Stochastic
Previous problem was deterministic: no new percepts
were necessary, we can predict the future perfectly given our actions
 Discrete / continuous
Previous problem was discrete: we can enumerate all possibilities
 Single Agent
No other agents interacting with your cost function
 Sequential
Decisions depend on past decisions 5
 Problem Formulation
A problem is defined by four items:

initial state e.g., "at Arad“

actions set of possible actions in current state x.

transition model Result(x,a) = state that follows from applying action

a in state x. e.g., Result(X2 (Arad), A4 (Arad  Zerind)) = X3 (Zerind)

goal test, e.g., x = "at Bucharest”, Checkmate(x)

path cost (additive)

 e.g., sum of distances, number of actions executed, etc.
 c(x,a,y) is the step cost, assumed to be ≥ 0

A solution is a sequence of actions leading from the initial state to a

goal state
6
 Selecting a state space
 Real world is absurdly complex
 state space must be abstracted for problem solving

 (Abstract) state  set of real states

 (Abstract) action  complex combination of real actions

 e.g., "Arad  Zerind" represents a complex set of possible routes, detours,
rest stops, etc.

 For guaranteed realizability, any real state "in Arad” must get to some
real state "in Zerind”

 (Abstract) solution  set of real paths that are solutions in the real world

 Each abstract action should be "easier" than the original problem

7
 Vacuum world state space graph

 states? discrete: dirt and robot location

 initial state? any
 actions? Left, Right, Suck
 goal test? no dirt at all locations
 path cost? 1 per action
8
 Example: 8-Queens

 states? -any arrangement of n<=8 queens

-or arrangements of n<=8 queens in leftmost n
columns, 1 per column, such that no queen
attacks any other.
 initial state? no queens on the board
 actions? -add queen to any empty square
-or add queen to leftmost empty square such
that it is not attacked by other queens.
 goal test? 8 queens on the board, none attacked.
 path cost? 1 per move 9
 Example: robotic assembly

 states?: real-valued coordinates of robot joint

angles parts of the object to be assembled
 initial state?: rest configuration
 actions?: continuous motions of robot joints
 goal test?: complete assembly
 path cost?: time to execute+energy used
10
 Example: The 8-puzzle

 states? locations of tiles

 initial state? given
 actions? move blank left, right, up, down
 goal test? goal state (given)
 path cost? 1 per move

[Note: optimal solution of n-Puzzle family is NP-hard] 11

 Tree search algorithms
 Basic idea:
 Exploration of state space by generating successors of
already-explored states (a.k.a.~expanding states).

 Every states is evaluated: is it a goal state?

12
Tree search example

13
Tree search example

14
Tree search example

15
 Repeated states
 Failure to detect repeated states can turn a
linear problem into an exponential one!

16
 Solutions to Repeated States
S
B
S
B C

C C S B S
State Space
Example of a Search Tree

 Graph search optimal but memory inefficient

 never generate a state generated before

 must keep track of all possible states (uses a lot of memory)
 e.g., 8-puzzle problem, we have 9! = 362,880 states

17
 Graph Search vs Tree Search

(see also blog entry: http://qaintroai.blogspot.in/2009/10/q-how-is-graph-search-different-from.html)

18
 Implementation: states vs. nodes
 A state is a (representation of) a physical configuration

 A node n is a data structure constituting part of a search

tree contains info such as: state, parent node, action, path
cost g(n).

 The Expand function creates new nodes, filling in the

various fields and using the SuccessorFn of the problem
to create the corresponding states. 19
 Search strategies
 A search strategy is defined by picking the order of node
expansion

 Strategies are evaluated along the following dimensions:

 completeness: does it always find a solution if one exists?
 time complexity: How long does it take to find a solution?
 space complexity: How much memory is needed to perform the
search?
 optimality: does it always find a least-cost solution?

 Time and space complexity are measured in terms of

 b: maximum branching factor of the search tree
 d: depth of the least-cost solution
 m: maximum depth of the state space (may be ∞)
20
Uninformed Search

21
 Complexity Recap
• We often want to characterize algorithms independent of their
implementation.

• “This algorithm took 1 hour and 43 seconds on my laptop”.

Is not very useful, because tomorrow computers are faster.

• Better is:
“This algorithm takes O(nlog(n)) time to run and O(n) to store”.
because this statement abstracts away from irrelevant details.

Time(n) = O(f(n)) means:

Time(n) < constant x f(n) for n>n0 for some n0
Space(n) –do--

n is some variable which characterizes the size of the problem,

e.g. number of data-points, number of dimensions, branching-factor
of search tree, etc.

• Worst case analysis versus average case analysis 22

 Uninformed search strategies
 Uninformed: While searching you have no clue
whether one non-goal state is better than any
other. Your search is blind. You don’t know if your
current exploration is likely to be fruitful.
Various blind strategies:
 Breadth-first search

 Uniform-cost search

 Depth-first search

 Iterative deepening search

23
 Breadth-first search
 Expand shallowest unexpanded node
 Fringe: nodes waiting in a queue to be explored
 Implementation:
 fringe is a first-in-first-out (FIFO) queue, i.e.,
new successors go at end of the queue.

Is A a goal state?

24
 Breadth-first search
 Expand shallowest unexpanded node
 Implementation:
 fringe is a FIFO queue, i.e., new successors go
at end

Expand:
fringe = [B,C]

Is B a goal state?

25
 Breadth-first search
 Expand shallowest unexpanded node


 Implementation:
 fringe is a FIFO queue, i.e., new successors go
at end
Expand:
fringe=[C,D,E]

Is C a goal state?

26
 Breadth-first search
 Expand shallowest unexpanded node
 Implementation:
 fringe is a FIFO queue, i.e., new successors go
at end

Expand:
fringe=[D,E,F,G]

Is D a goal state?

27
Example
BFS

28
 Properties of breadth-first search
 Complete? Yes it always reaches goal (if b is finite)
 Time? 1+b+b2+b3+… +bd + (bd+1-b)) = O(bd+1)
(this is the number of nodes we generate)
 Space? O(bd+1) (keeps every node in memory,
either in fringe or on a path to fringe).
 Optimal? Yes (if we guarantee that deeper
solutions are less optimal, e.g. step-cost=1).

 Space is the bigger problem (more than time)

Note: in the new edition Space & Time complexity was O(bd) because we
postpone the expansion. 29
 Uniform-cost search
Breadth-first is only optimal if step costs is increasing with
depth (e.g. constant). Can we guarantee optimality for any
step cost?
Uniform-cost Search: Expand node with
smallest path cost g(n).
Proof Completeness:

Given that every step will cost more than 0,

and assuming a finite branching factor, there
is a finite number of expansions required before
the total path cost is equal to the path cost of the
goal state. Hence, we will reach it.

Proof of optimality given completeness:

Assume UCS is not optimal.

Then there must be an (optimal) goal state with
path cost smaller than the found (suboptimal) goal
state (invoking completeness).
However, this is impossible because UCS would
have expanded that node first by definition.
Contradiction.

30
 Uniform-cost search

Implementation: fringe = queue ordered by path cost

Equivalent to breadth-first if all step costs all equal.

Complete? Yes, if step cost ≥ ε

(otherwise it can get stuck in infinite loops)

Time? # of nodes with path cost ≤ cost of optimal solution.

Space? # of nodes with path cost ≤ cost of optimal solution.

Optimal? Yes, for any step cost ≥ ε

31
 6 1
3 A D F 1

2 4 8
S B E G

1 20
C

The graph above shows the step-costs for different paths going from the start (S) to
the goal (G).

Use uniform cost search to find the optimal path to the goal.

Exercise

32
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = Last In First Out (LIFO) queue, i.e., put
successors at front

Is A a goal state?

33
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[B,C]

Is B a goal state?

34
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[D,E,C]

Is D = goal state?

35
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[H,I,E,C]

Is H = goal state?

36
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[I,E,C]

Is I = goal state?

37
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[E,C]

Is E = goal state?

38
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[J,K,C]

Is J = goal state?

39
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[K,C]

Is K = goal state?

40
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[C]

Is C = goal state?

41
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[F,G]

Is F = goal state?

42
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[L,M,G]

Is L = goal state?

43
 Depth-first search
 Expand deepest unexpanded node
 Implementation:
 fringe = LIFO queue, i.e., put successors at front

queue=[M,G]

Is M = goal state?

44
 Properties of depth-first search
A

 Complete? No: fails in infinite-depth spaces B C

Can modify to avoid repeated states along path

 Time? O(bm) with m=maximum depth
 terrible if m is much larger than d
 but if solutions are dense, may be much faster than
breadth-first
 Space? O(bm), i.e., linear space! (we only need to
remember a single path + expanded unexplored nodes)
 Optimal? No (It may find a non-optimal goal first)

45
 Iterative deepening search

• To avoid the infinite depth problem of DFS, we can

decide to only search until depth L, i.e. we don’t expand beyond depth L.
 Depth-Limited Search

• What if solution is deeper than L?  Increase L iteratively.

 Iterative Deepening Search

• As we shall see: this inherits the memory advantage of Depth-First

search, and is better in terms of time complexity than Breadth first search.

46
Iterative deepening search L=0

47
Iterative deepening search L=1

48
Iterative deepening search L=2

49
Iterative Deepening Search L=3

50
 Iterative deepening search
 Number of nodes generated in a depth-limited search to
depth d with branching factor b:
NDLS = b0 + b1 + b2 + … + bd-2 + bd-1 + bd

 Number of nodes generated in an iterative deepening

search to depth d with branching factor b:
NIDS = (d+1)b0 + d b1 + (d-1)b2 + … + 3bd-2 +2bd-1 + 1bd =
O (b d )  O (b d 1 )

 For b = 10, d = 5, BFS

 NDLS = 1 + 10 + 100 + 1,000 + 10,000 + 100,000 = 111,111
 NIDS = 6 + 50 + 400 + 3,000 + 20,000 + 100,000 = 123,450
 NBFS = ............................................................................................ = 1,111,100
d
Note: BFS can also be adapted to be O (b ) by waiting to expand until all nodes at depth d are checked
51
 Properties of iterative deepening search

 Complete? Yes
 Time? O(bd)
 Space? O(bd)
 Optimal? Yes, if step cost = 1 or increasing
function of depth.

52
 Bidirectional Search
 Idea
 simultaneously search forward from S and backwards
from G
 stop when both “meet in the middle”
 need to keep track of the intersection of 2 open sets of
nodes
 What does searching backwards from G mean
 need a way to specify the predecessors of G
 this can be difficult,
 e.g., predecessors of checkmate in chess?
 which to take if there are multiple goal states?
 where to start if there is only a goal test, no explicit list?53
Bi-Directional Search
Complexity: time and space complexity are:
O (b d /2 )

54
Summary of algorithms

55
 Summary
 Problem formulation usually requires abstracting away real-
world details to define a state space that can feasibly be
explored

 Variety of uninformed search strategies

 Iterative deepening search uses only linear space and not

much more time than other uninformed algorithms

http://www.cs.rmit.edu.au/AI-Search/Product/
http://aima.cs.berkeley.edu/demos.html (for more demos)

56