ARTIFICIAL INTELLIGENCE

Course Code: CS-3811

 Constraint
Satisfaction Problems

2
 Constraint Satisfaction
Problems
 constraint satisfaction is problem solving technique. It is a finite choices decision
problem. Where one is given a fixed set of decisions to make .
 Each decision involves choosing among a fixed set of options. Each constraint restricts
the combination of choices that can be taken simultaneously.
 The task is to make all decisions such that all the constraints are satisfied.
 Many real life problems can be dived by using CSP. For example timetabling ,
planning .
 Following are CSP's
 ● cryptarithemetic ● The n-queens problem
 ● A crossword problem ● A Map Coloring problem
 ● Latin Square problem ● 8 queen puzzle problem
 ● Sudoku problem
CONT.

 constraint satisfaction problems, whose states and goal

test conform to a standard, structured, and very simple
representation Search algorithms can be defined that
take advantage of the structure of states and use
general-purpose rather than problem-specific heuristics
to enable the solution of large problems
 Formally speaking, a constraint satisfaction problem (or CSP) is defined
by a set of VARIABLES XI, X2,. . . , Xn, and a set of constraints, C1, (72,.
. . , C,. Each variable Xi has a nonempty domain Di of possible values.
Each constraint Ci involves some subset of the variables and specifies the
allowable combinations of values for that subset. A state of the problem
is defined by an assignment of values to some or all of the variables, {Xi
= vi, Xj = vj, . . .). An assignment that does not violate any constraints is
called a consistent or legal assignment. A complete assignment is one in
which every variable is mentioned, and a to a CSP is a complete
assignment that satisfies all the constraints. Some CSPs also require a
solution that maximizes an objective function.

5
Constraint satisfaction problems
(CSPs)

 CSP:
 state is defined by variables Xi with values
from domain Di
 goal test is a set of constraints specifying
allowable combinations of values for subsets of
variables

 Allows useful general-purpose algorithms

with more power than standard search
algorithms
6
 Example: Map-Coloring

 Variables WA, NT, Q, NSW, V, SA, T

 Domains Di = {red,green,blue}

 Constraints: adjacent regions must have different colors

 e.g., WA ≠ NT
7
 Example: Map-Coloring

 Solutionsare complete and consistent

assignments, e.g., WA = red, NT = green,Q =
red,NSW = green,V = red,SA = blue,T = green
8
Constraint graph

 Binary CSP: each constraint relates two

variables
 Constraint graph: nodes are variables,
arcs are constraints

9
Varieties of CSPs

 Discrete variables
 finite domains:
 n variables, domain size d  O(d n) complete assignments
 e.g., 3-SAT (NP-complete)
 infinite domains:
 integers, strings, etc.
 e.g., job scheduling, variables are start/end days for each job:
StartJob1 + 5 ≤ StartJob3

 Continuous variables
 e.g., start/end times for Hubble Space Telescope observations
 linear constraints solvable in polynomial time by linear
programming

10
Varieties of constraints

 Unary constraints involve a single variable,

 e.g., SA ≠ green

 Binary constraints involve pairs of variables,

 e.g., SA ≠ WA

 Higher-order constraints involve 3 or more

variables,
 e.g., SA ≠ WA ≠ NT

11
 Example: Cryptarithmetic

 Variables: F T U W R O X1 X2 X3
 Domains: {0,1,2,3,4,5,6,7,8,9} {0,1}
 Constraints: Alldiff (F,T,U,W,R,O)
 O + O = R + 10 · X1
 X1 + W + W = U + 10 · X2
 X2 + T + T = O + 10 · X3 12

 X3 = F, T ≠ 0, F ≠ 0
 Real-world CSPs
 Assignment problems
 e.g., who teaches what class
 Timetabling problems
 e.g., which class is offered when and where?
 Transportation scheduling
 Factory scheduling

 Notice that many real-world problems involve

real-valued variables

13
 Standard search formulation
Let’s try the standard search formulation.

We need:
• Initial state: none of the variables has a value (color)
• Successor state: one of the variables without a value will get some value.
• Goal: all variables have a value and none of the constraints is violated.

NxD
N layers
WA WA WA NT T
[NxD]x[(N-1)xD]
WA WA WA NT
NT NT NT WA

Equal! 14
N! x D^N
There are N! x D^N nodes in the tree but only D^N distinct states??
Backtracking (Depth-First)
search
•Special property of CSPs: They are commutative:
NT = WA
This means: the order in which we assign variables
WA NT
does not matter.

• Better search tree: First order variables, then assign them values one-by-one.

D
WA WA WA
WA
NT D^2
WA WA
NT NT

D^N 15
Backtracking example

16
Backtracking example

17
Backtracking example

18
Backtracking example

19
Improving backtracking
efficiency
 General-purpose methods can give huge gains in speed:
 Which variable should be assigned next?
 In what order should its values be tried?
 Can we detect inevitable failure early?

 We’ll discuss heuristics for all these questions in the

following.

20
 Which variable should be assigned next?
minimum remaining values heuristic

 Most constrained variable:

choose the variable with the fewest legal values

 a.k.a. minimum remaining values (MRV) heuristic

 Picks a variable which will cause failure as soon as
possible, allowing the tree to be pruned.

21
 Which variable should be assigned next? 
degree heuristic

 Tie-breaker among most constrained variables

 Most constraining variable:

 choose the variable with the most constraints on remaining
variables (most edges in graph)

22
 In what order should its values be tried?
 least constraining value heuristic

 Given a variable, choose the least constraining value:

 the one that rules out the fewest values in the remaining
variables

 Leaves maximal flexibility for a solution.

 Combining these heuristics makes 1000 queens feasible

23
Rationale for MRV, DH, LCV

 In all cases we want to enter the most promising branch,

but we also want to detect inevitable failure as soon as
possible.

 MRV+DH: the variable that is most likely to cause failure in

a branch is assigned first. The variable must be assigned at
some point, so if it is doomed to fail, we’d better found
out soon. E.g X1-X2-X3, values 0,1, neighbors cannot be the
same.

 LCV: tries to avoid failure by assigning values that leave

maximal flexibility for the remaining variables. We want
our search to succeed as soon as possible, so given some
ordering, we want to find the successful branch.
24
 Can we detect inevitable failure early?
 forward checking

 Idea:
 Keep track of remaining legal values for
unassigned variables
that are connected to current variable.
 Terminate search when any variable has no
legal values

25
Forward checking

 Idea:
 Keep track of remaining legal values for
unassigned variables
 Terminate search when any variable has no
legal values

26
Forward checking

 Idea:
 Keep track of remaining legal values for
unassigned variables
 Terminate search when any variable has no
legal values

27
Forward checking

 Idea:
 Keep track of remaining legal values for
unassigned variables
 Terminate search when any variable has no
legal values

28
2) Consider the constraint graph on the right.
The domain for every variable is [1,2,3,4].
There are 2 unary constraints:
- variable “a” cannot take values 3 and 4.
- variable “b” cannot take value 4.
There are 8 binary constraints stating that variables
connected by an edge cannot have the same value.
b
a) [5pts] Find a solution for this CSP by using the following
heuristics: minimum value heuristic, degree heuristic,
forward checking. Explain each step of your answer.
a c e

d
29

CONSTRAINT GRAPH
2) Consider the constraint graph on the right.
The domain for every variable is [1,2,3,4].
There are 2 unary constraints:
- variable “a” cannot take values 3 and 4.
- variable “b” cannot take value 4.
There are 8 binary constraints stating that variables
connected by an edge cannot have the same value.
b
a) [5pts] Find a solution for this CSP by using the following
heuristics: minimum value heuristic, degree heuristic,
forward checking. Explain each step of your answer.
a c e
MVH a=1 (for example)
FC+MVH b=2
FC+MVH+DH c=3
FC+MVH d=4
d
FC e=1
30

CONSTRAINT GRAPH
Constraint propagation

 Forward checking only checks consistency between

assigned and non-assigned states. How about constraints
between two unassigned states?

 NT and SA cannot both be blue!

 Constraint propagation repeatedly enforces constraints
locally

31
Arc consistency

 Simplest form of propagation makes each

arc consistent
 X Y is consistent iff
for every value x of X there is some allowed y

consistent arc.

constraint propagation propagates arc consistency on the 32graph.

Arc consistency

 Simplest form of propagation makes each

arc consistent
 X Y is consistent iff
for every value x of X there is some allowed y

inconsistent arc.
remove blue from source consistent arc.

33
Arc consistency

 Simplest form of propagation makes each arc consistent

 X Y is consistent iff
for every value x of X there is some allowed y

 this arc just became inconsistent

If X loses a value, neighbors of X need to be rechecked:
i.e. incoming arcs can become inconsistent again
(outgoing arcs will stay consistent).

34
 Arc consistency

 Simplest form of propagation makes each arc consistent

 X Y is consistent iff
for every value x of X there is some allowed y

 If X loses a value, neighbors of X need to be rechecked

 Arc consistency detects failure earlier than forward checking
 Can be run as a preprocessor or after each assignment

# arcs 35

 Time complexity: O(n2d3) d^2 for checking, each node can be checked d times at most
 Arc Consistency

 This is a propagation algorithm. It’s like sending messages to

neighbors on the graph! How do we schedule these messages?

 Every time a domain changes, all incoming messages need to be re-

send. Repeat until convergence  no message will change any
domains.

 Since we only remove values from domains when they can never be
part of a solution, an empty domain means no solution possible at all
 back out of that branch.

 Forward checking is simply sending messages into a variable

36 that just
got its value assigned. First step of arc-consistency.
Try it yourself
[R,B,G] [R,B,G]

[R]

[R,B,G] [R,B,G]

Use all heuristics including arc-propagation to solve this problem.

37
38
 B G R R G B

B a priori
B B R G B
R constrained
G R
G G
nodes

Note: After the backward pass, there is guaranteed

to be a legal choice for a child node for any of its
leftover values.
This removes any inconsistent values from Parent(Xj),
it applies arc-consistency moving backwards.
39
40
Junction Tree Decompositions

41
Local search for CSPs

 Note: The path to the solution is unimportant, so we can

apply local search!

 To apply to CSPs:
 allow states with unsatisfied constraints
 operators reassign variable values

 Variable selection: randomly select any conflicted variable

 Value selection by min-conflicts heuristic:

 choose value that violates the fewest constraints
 i.e., hill-climb with h(n) = total number of violated constraints

42
Example: 4-Queens

 States: 4 queens in 4 columns (44 = 256

states)
 Actions: move queen in column
 Goal test: no attacks
 Evaluation: h(n) = number of attacks

43
44
45
46
47
48
49
50
51
52
53
Summary

 CSPs are a special kind of problem:

 states defined by values of a fixed set of variables
 goal test defined by constraints on variable values

 Backtracking = depth-first search with one variable assigned per

node

 Variable ordering and value selection heuristics help significantly

 Forward checking prevents assignments that guarantee later failure

 Constraint propagation (e.g., arc consistency) does additional work

to constrain values and detect inconsistencies

 Iterative min-conflicts is usually effective in practice

54