Unit-3

Constraint Satisfaction Problems

Chapter 6

Book: Artificial Intelligence (Stuart Russel and Peter Norvig)

4th edition

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Introduction to CSP
• CSP is defined by a set of variables, X1, X2, ....Xn, and
a set of constraints C1, C2,...Cm.
• Each variable Xi has a non-empty domain Di of
possible values.
• Each constraint Ci involves some subset of the
variables and specifies the allowable combination 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,...}

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Introduction to CSP (terminology)
• An assignment that does not violate any constraint is
called a consistent or a legal assignment.
• A complete assignment is one in which every
variable is mentioned and a solution to CSP is a
complete assignment that satisfies all the
constraints.
• Some CSPs also require a solution that maximizes the
objective function.
• Constraint satisfaction problems (CSPs) consist of
variables with constraints on them. The structure of
a CSP can be represented by a constraint graph.
Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
The map colouring problem

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
The map colouring problem as a CSP

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Varied types of problems as a CSP
• Many kinds of design and scheduling problems can be
expressed as CSPs, so they form a very important subclass.
CSPs can be solved by general-purpose search algorithms, but
because of their special structure, algorithms designed
specifically for CSPs generally perform much better.
• For example, the 8-queens problem can be viewed as a CSP in
which the variables are the locations of each of the eight
queens; the possible values are squares on the board; and the
constraints state that no two queens can be in the same row,
column or diagonal.
• A solution to a CSP specifies values for all the variables such
that the constraints are satisfied.
• Cryptarithmetic and VLSI layout can also be described as CSPs
. Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
 Approaches to solving CSPs
• Every solution of CSP must be a complete assignment
and therefore appears at depth n, if there are n
variables. The search tree also extends till depth n.
• Therefore, depth first search algorithms are popular
for solving CSPs.
• Moreover, the path by which the solution is reached
is irrelevant. Due to this local search methods are
also suitable for solving CSPs. These methods use
complete-state formulation in which every state is a
complete assignment that might or might not satisfy
the constraints.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 CSP variables, domains and constraints
• CSP variables can be discrete and have finite
domains.
– Map coloring problems
• If the maximum domain size of any variable is d, then
the number of possible complete assignments is
O(dn). Here n is the number of variables
• Finite domain CSPs include Boolean CSPs, whose
variables can be either true or false.
• Discrete variables can also have infinite domains.
– Set of integers
– infinite domains require a constraint language to describe
constraints
Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
 Type of constraints

• Unary constraints: They restrict the value of a single

variable
• Binary constraints: They relate two variables
– WA ≠ NT
• Higher order constraints involves three or more
variables.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 A cryptarithmetic problem
• Each letter stands for a distinct digit.
• The aim is to find a substitution of digits for letters
such that the resulting sum is arithmetically correct.
• No leading zeros are allowed.
– a) Puzzle b) Constraint Graph for the puzzle

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Solving cryptarithmetic puzzle as a CSP
TWO + TWO = FOUR
• Identifying the variables and domain
Variables: T, W, O, F, U, R, These variables belong to the
domain {0,1,2,3,4,5,6,7,8,9}
• Identifying the constraints:
– F ≠ T, F ≠ W and so on (for all variables)
– Addition constraints on the four columns:
• O+O=R+10*X1
• X1+W+W=U+10*X2
• X2+T+T=O+10*X3
• X3=F
• X1,X2,X3 are auxiliary variables representing the digits(0 or 1,
carried over to the next column)

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Cryptarithmetic Solution
X3 can be (0,1), leading digits cannot be zero, so X3=1, and hence F=1.
X2+T+T=O+10*X3, X3=1, X2=(0,1), assume X2=0 for some solution
Therefore T+T=O+10.
Starting from T=(0,1,….9), T=1,2,3,4 not possible, 5 not possible as O will be
zero and O+O=R is violated. Therefore legal values of T= [6,7,8,9]
Let T=6, O=2, O+O=2+2=4, therefore R=4, so X1=0, and X2=0 (already
assumed), so W can take the values (1,2,3,4) only.
But F=1, O=2, T=6(As W+W=U, 3+3=6, so U=6 not possible, so W=3 not
possible), R=4.
Thus, W cannot take the values (1,2,3,4), so no legal values available hence
backtrack to T=6 and choose a new value for T, choose T=7
Therefore 7+7=O+10, O=4, O+O=R+10*X1, so X1=0 and R=8, W=3, U=6
Thus, all variables are assigned, and all constraints are satisfied.
Solution: T=7, W=3, O=4, F=1, U=6, R=8, 7 3 4
7 3 4
1 4 6 8
Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
 Backtracking search for CSPs
• Backtracking search is a kind of depth first search
which is commonly used for solving CSPs.
• It chooses values for one variable at a time and
backtracks when a variable has no legal values left to
assign.
• Plain backtracking is an uninformed algorithm.
• The search efficiency of the Backtracking search can
be improved by ordering of variables as well as
ordering of variable values in a specific way.
• This is called variable and value ordering in CSPs. It
acts as a kind of heuristics.
Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
 Backtracking search for CSPs
(with heuristics)

• The minimum remaining values and degree

heuristics are methods for deciding which variable to
choose next in a backtracking search.
• The least constraining value heuristic helps in
ordering the variable values.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Variable ordering (MRV)
Minimum remaining value (MRV)heuristic: It is also
called most constrained variable or fail first heuristic.
MRV chooses the variable with the fewest legal values.
It is called Fail first as it picks a variable that is most
likely to cause a failure.
Advantages:
– It prunes the search tree.
– If there is some variable X with zero legal values remaining,
MRV will pick it first and failure will be detected
immediately. This avoids pointless search through other
variables, as the search s bound to be a failure later-on
when X is selected.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Variable ordering (degree heuristics)

• The degree heuristics attempts to reduce the

branching factor by selecting the variable that is
involved in largest number of constraints on other
unassigned variables.
• In the map coloring algorithm, the degree of SA is 5
(highest). Once SA is chosen by applying degree
heuristics, the problem can be solved without any
false steps and arrive at a solution without back
tracking.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Least constraining value heuristic(value ordering)

• Once a variable has been selected, the algorithm

must decide on the order in which the values should
be examined.
• Least constraining value (LCV) heuristic prefers that
value of the variable which does not reduce the
choices of other constrained variables.
• For example: let the partial assignment be {WA=red,
NSW=green}, next choice is to be done for variable Q.
If Q is assigned blue, it would be a bad choice as it
eliminates the last legal value left for SA (SA cannot
be red or green). The LCV therefore prefers red
instead of blue.
Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
 Propagating information through constraints
• If some constraints are looked at earlier in the
search, then it can reduce the search space
considerably.
• This can be done by forward checking process.
• Whenever a variable X is assigned, the forward
checking process looks at each unassigned variable Y
that is connected to X by a constraint. It deletes from
Y’s domain, any value that is inconsistent with the
value chosen for X.
• The forward checking can detect that the partial
assignment done till now is inconsistent.
Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
Applying forward checking process

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Forward checking process
• Consider the color assignment sequence as WA, Q, V, NT.
• After assigning WA = red and Q = green, and V= blue,the
domains of NT and SA are reduced to a single value; the
branching on these variables has been eliminated altogether
by propagating information from WA and Q.
• The MRV heuristic, which is used for forward checking, would
automatically select SA and NT next.
• A second point to notice is that, after V = blue, the domain of
SA is empty. Hence, forward checking has detected that the
partial assignment { WA = red, Q = green, V = blue) is
inconsistent with the constraints of the problem, and the
algorithm will therefore backtrack immediately.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Intelligent Backtracking Search for CSP
Intelligent backtracking: looking backwards
• In simple backtracking, when a branch of search fails, the
algorithm backs up to the preceding variable and try a
different value for it. This is called chronological backtracking.
– Backtracking goes back one level in the search tree at a time.
– Not rational in cases where the previous step is not involved with the
conflict
• Consider fixed variable ordering Q, NSW, V, T, SA, WA, NT.
Suppose partial assignment is as follows:
• {Q= red, NSW= green, V= blue, T=red}
• When next variable SA is taken up for assignment, no valid
colour is available(). Backtracking to T does not help as T’s
value does not affect other nodes’ constraint.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Backjumping
• Backtracking needs to be done till one arrives at a set
of variables from which one variable caused the
failure. This set is called the conflict set. In this case
conflict set is {Q, NSW, V}.
• Definition: Conflict set for a variable X is the set of
previously assigned variables that are connected to
X by constraints.
• The backjumping method backtracks to the most
recent variables in the conflict set.
• In this case backjumping jumps over T and tries a
new value for V.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Backjumping contd..
• To enable backjumping the conflict set needs to be
maintained while checking for a legal value to assign. If
no legal value is found the most recent element of the
conflict set is returned with failure indicator.
• Simple backjumping detects when every value in domain
is in conflict with current assignment. But if forward
checking mechanism is used then such a state is never
reached at all. Every branch pruned by basic backjumping
is also pruned by forward checking
• Due to this simple backjumping becomes redundant.
Therefore conflict-directed backjumping is preferred.

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Conflict-directed backjumping
• Backjumping notices failure when a variable’s
domain becomes empty, but in many cases a branch
has failed long before actual failure occurs.
• Consider the partial assignment {WA=red, NSW=red}
• Suppose we set T=red next. Then we assign NT, Q, V
and SA. {WA, NSW, T, NT, Q, V, SA}
• No valid assignment can work for these. Due to this
we are left with no values to try out at NT. See figure
SA=Blue, conflicts with V
SA= Red or Green conflicts
with WA and Q.
• Where to backtrack now?
Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
 Conflict-directed backjumping (CBJ) contd..
• It is a fact that NT, Q, V, SA, taken together, failed
because of a set of preceding variables. These must
be the variables that directly conflict with the
variables NT, Q, V, SA.
• Deeper notion of conflict set: It is the set of
preceding variables (WA and NSW) that caused NT,
together with any subsequent variables, to have no
consistent solution. As per CBJ the conflict set is {WA
and NSW}. The algorithm should backtrack to NSW
(as it is recent) and skip over Tasmania (T).
• A back-jumping algorithm that uses conflict sets in
this way is called as conflict-directed backjumping.
Artificial Intelligence Slides-- Aparna Gurjar
CSE(AIML)
 CBJ definition
• Conflict-directed backjumping (CBJ):
– Let Xj be the current variable, and conf(Xj) its
conflict set. If every possible value for Xj fails,
backjump to the most recently assigned variable
in conf(Xj), denoted by Xi.
– Set conflict set of Xi to
conf(Xi) = conf(Xi)  conf(Xj) – {Xi}
– If we need to backjump from Xi, repeat this
process.
View example in next slide

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)
 Conflict-directed backjumping example
SA fails, SA conflict set is {WA, NT, Q}, we back-jump to Q and Q
absorbs the conflict set from SA into its own conflict set. Q
conflict set is {NT, NSW}. Q absorbs the conflict set from SA
minus itself.
{WA, NT, Q} U {NT, NSW} – {Q}. The new conflict set of Q is {WA,
NT, NSW} and so on.
Now Q fails(as there is no solution from Q onwards), therefore
backjump to NT (which is most recent in Q’s conflict set). NT
absorbs {WA, NT, NSW} from Q. Its own conflict set is {WA}.
{WA} U {WA, NT, NSW} – {NT}. The new conflict set of NT
becomes {WA, NSW}.
NT still fails, back-jump to NSW (which is most recent in NT’s
conflict set). Thus, from SA the algorithm finally backtracks to
NSW, skipping over Tasmania (T). {WA, NSW, T, NT, Q, V, SA}

Artificial Intelligence Slides-- Aparna Gurjar

CSE(AIML)