Heuristic Functions

How to evaluate them?

How to find them?
Admissibility
Examples
search

Heuristic function
 A function which applies an algorithm to
the nodes and assigns a value to them that
indicates how close the evaluated node is
to the goal node.
 Difficulty however is in determining the
heuristics
 Given a heuristic guidance is still required
as to how to use the heuristic values
2
search

Finding Heuristics
Tic - Tac - Toe
which one to choose? X

X
X
Heuristic:
calculate winning lines and move to state with most wining lines.
Note that here we have a maximizing heuristic, maximum is
chosen, otherwise minimum is chosen for minimizing heuristic
3
search

Maximum value is chosen

Maximizing Minimizing

Minimum value is chosen

Heuristic Heuristic
(choosing a (distance from
move that the goal)
increases wining
chances)

4
search

Calculating winning lines

X X
X

3 winning 4 winning 3 winning

lines lines lines

Always choose the state with maximum heuristic value

Maximizing heuristics

5
search

Example TSP

Choosing city with minimum distance from the goal:

Always choose the city with minimum heuristic value
Minimizing heuristics (hi<hi-1)
6
search

8 Queen Problem

 8-queens problem (complete-state

formulation).
 Successor function: move a single queen to
another square in the same column.
7
search

Min-conflicts Example 1

h=5 h=3 h=1

min-conflicts heuristic function h(n) the

Count the number of queens attacking a

given queen in a row or a column.
8
 Min-conflicts Example 2 (Grid
search

Score)

 A two-step solution for an 8-queens problem using min-conflicts

heuristic

 At each stage a queen is chosen for reassignment in its column

 The algorithm moves the queen to the min-conflict square breaking

ties randomly. 9
search

8 Puzzle Problem: Heuristic functions

 8-puzzle
 Avg. solution cost is about 22 steps (branching factor
+/- 3)
 Exhaustive search to depth 22: 3.1 x 1010 states.
 A good heuristic function can reduce the search process.
 CAN YOU THINK OF A HEURISTIC ?

10
search

8 Puzzle

 Two commonly used heuristics

 h1 = the number of misplaced tiles
 h1(s)=8
 h2 = the sum of the distances of the tiles from
their goal positions (manhattan distance).
 h2(s)=3+1+2+2+2+3+3+2=18

11
search

Heuristic quality

 Effective branching factor b*

 Is the branching factor that a uniform tree of depth
d would have in order to contain N+1 nodes.

N  1  1 b * (b*) 2  ... (b*) d

 A good heuristic should have b* as low as possible
 This measure is fairly constant for sufficiently hard
problems.
  Can thus provide a good guide to the heuristic’s overall
usefulness.
 A good value of b* is 1.

12
search

Heuristic quality and dominance

h1 h2 h1 h2

 Effective branching factor of h2 is lower than h1

 If h2(n) >= h1(n) for all n then h2 dominates h1 and is
better for search (value of h2 and h1 e.g. 18 vs 8)
13
search

Heuristic Function: Costs

 Applying heuristic function reduces the
branching factor, thus reduces the
computational cost
 The cost of applying the heuristic is
much less than the cost of evaluating
the complete search tree

14
search

Admissibility: 8 Puzzle

 Two commonly used heuristics

 h1 = the number of misplaced tiles h1(s)=8
 h2 = the sum of the distances of the tiles from
their goal positions (manhattan distance).
 h2(s)=3+1+2+2+2+3+3+2=18
• What is the actual number of moves required to solve this problem
• For tile ‘1’ alone almost 14 moves are required
• This is higher than total value for h1 or h2 (for all tiles)
• So h1 and h2 under estimate the actual cost. 15
search

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

16
search

Why not Overestimate ?

May take the search

away from goal

17
search

Which Heuristic is Admissible?

 h1 = the number of misplaced tiles

 h1(s)=8
 h2 = the sum of the distances of the tiles from their
goal positions (manhattan distance).
 h (s)=3+1+2+2+2+3+3+2=18
2

 BOTH
18
search

Manhattan Distance: is it Admissible

3 1 1 3

Manhattan distance is 2+2=4 moves

19
search

Manhattan Distance: is it Admissible

3 1 1 3

Manhattan distance is 2+2=4 moves

20
search

Manhattan Distance: is it Admissible

3 1 3
1

Manhattan distance is 2+2=4 moves

21
search

Manhattan Distance: is it Admissible

3 1 3
1

Manhattan distance is 2+2=4 moves

22
search

Manhattan Distance: is it Admissible

3 1 3
1

Manhattan distance is 2+2=4 moves

23
search

Manhattan Distance: is it Admissible

1 3 1 3

Manhattan distance is 2+2=4 moves

24
search

Manhattan Distance: is it Admissible

1 3 1 3

Manhattan distance is 2+2=4 moves,

But actually 2 additional moves are required. 25
search

B
Consider the state-space on the
1 5
Figure.
S G

Provide an analysis of whether the 2 4

following heuristic functions are A
admissible or not for this state-
space. 3

S A B G
h1: 0 0 0 0
h2: 2 3 5 0
h3: 3 4 3 0
h4: 4 0 0 0

ANS: h* (S) = 3, h*(A) = 4, h*(B) = 5, h*(G) = 0.

Admissible = {h1, h2, h3}
26
search

Finding admissible heuristics

 Relaxed Problem:
 Admissible heuristics can be derived from the exact
solution cost of a relaxed version of the problem:
 Relaxed 8-puzzle for h1 : a tile can move anywhere As a
result, h1(n) gives the shortest solution
 Relaxed 8-puzzle for h2 : a tile can move to any
adjacent square.
 As a result, h2(n) gives the shortest solution.

The optimal solution cost of a relaxed problem

is no greater than the optimal solution cost of
the real problem.
27
search

Example: Relaxed Problem

5 8 1 2 3

4 2 1 x x x

7 3 6 x x

STATE(N) GOAL STATE

Goal contains partial information, the sequence in the

first row is important.
28
search
Relaxed Problem: Example

5 8 1 2 3
4 2 1 4 5 6 8

7 3 6 7 8

5 6
6

h = h8 + h5 + h6 +…
29
search
Complex: Relaxed Problem
 Consider two more complex relaxed problems:
5 8 1 2 3

4 2 1 4 5 6

7 3 6 7 8

1 2 3 5 8

4 2 1 4 5 6

3 7 6 7 8

 h = d1234 + d5678 [disjoint pattern heuristic]

 These distances could have been pre-computed in a
database 30
search

Summary: Admissible Heuristic

 Defining and Solving Sub-problem:

 Admissible heuristics can also be derived from the
solution cost of a sub-problem of a given problem.
 This cost is a lower bound on the cost of the real
problem.
 Pattern databases store the exact solution for every
possible sub-problem instance.
 The complete heuristic is constructed using the
patterns in the DB
 Learning through experience
31