Constraint Satisfaction Problems

Chapter 6

TB Artificial Intelligence

Slides from AIMA — http://aima.cs.berkeley.edu

1 / 31
Outline

I CSP examples
I Backtracking search for CSPs
I Problem structure and problem decomposition
I Local search for CSPs

2 / 31
Constraint satisfaction problems (CSPs)

I Standard search problem: state is a “black box”—any old data structure that supports goal
test, eval, successor
I CSP:
I state is defined by variables Xi
I with values from domain Di
I goal test is a set of constraints specifying allowable combinations of values for subsets of
variables
I Simple example of a formal representation language
I Allows useful general-purpose algorithms with more power
than standard search algorithms

3 / 31
Example: Map-Coloring

Northern
Territory
Western Queensland
Australia

South
Australia
New South Wales

Victoria

I Variables WA, NT , Q, NSW , V , SA, T

Tasmania
I Domains Di = {red, green, blue}
I Constraints: adjacent regions must have different colors, e.g., WA 6= NT (if the language
allows this), or (WA, NT ) ∈ {(red, green), (red, blue), (green, red), (green, blue), . . .}

4 / 31
Example: Map-Coloring contd.

Northern
Territory
Western Queensland
Australia

South
Australia
New South Wales

Victoria

Tasmania

I Solutions are assignments satisfying all constraints, e.g.,

{WA = red, NT = green, Q = red, NSW = green,
V = red, SA = blue, T = green}
5 / 31
Constraint graph

I Binary CSP: each constraint relates at most two variables

I Constraint graph: nodes are variables, arcs show constraints

NT
Q
WA

SA NSW

V
Victoria

General-purpose CSP algorithms use the graph structure to speed up search. E.g., Tasmania is
an independent subproblem!

6 / 31
Varieties of CSPs

I Discrete variables
I finite domains; size d =⇒ O(d n ) complete assignments
I e.g., Boolean CSPs, incl. Boolean satisfiability (NP-complete)
I infinite domains (integers, strings, etc.)
I e.g., job scheduling, variables are start/end days for each job
I need a constraint language, e.g., StartJob1 + 5 ≤ StartJob3
I linear constraints solvable, nonlinear undecidable
I Continuous variables
I e.g., start/end times for Hubble Telescope observations
I linear constraints solvable in poly time by LP methods

7 / 31
Varieties of constraints

I Unary constraints involve a single variable

e.g., SA 6= green
I Binary constraints involve pairs of variables
e.g., SA 6= WA
I Higher-order constraints involve 3 or more variables
e.g., cryptarithmetic column constraints
I Preferences (soft constraints)
e.g., red is better than green
often representable by a cost for each variable assignment
→ constrained optimization problems

8 / 31
Example: Cryptarithmetic

T WO F T U W R O
+ T WO
F O U R
X3 X2 X1

I Variables: F T U W R O X1 X2 X3
I Domains: {0, 1, 2, 3, 4, 5, 6, 7, 8, 9}
I Constraints
alldiff(F , T , U, W , R, O)
O + O = R + 10 · X1
...
9 / 31
Real-world CSPs

I Assignment problems
e.g., who teaches what class
I Timetabling problems
e.g., which class is offered when and where?
I Hardware configuration
I Spreadsheets
I Transportation scheduling
I Factory scheduling
I Floorplanning
I ...

Notice that many real-world problems involve real-valued variables

10 / 31
Standard search formulation (incremental)

Let’s start with the straightforward, dumb approach, then fix it

States are defined by the values assigned so far

I Initial state: the empty assignment, { }

I Successor function: assign a value to an unassigned variable that does not conflict with
current assignment =⇒ fail if no legal assignments (not fixable!)
I Goal test: the current assignment is complete

1. This is the same for all CSPs!

2. Every solution appears at depth n with n variables =⇒ use depth-first search
3. Path is irrelevant, so can also use complete-state formulation
4. b = (n − `)d at depth `, hence n!d n leaves!!!!

11 / 31
Backtracking search

I Variable assignments are commutative, i.e.,

[WA = red then NT = green] same as [NT = green then WA = red]
I Only need to consider assignments to a single variable at each node
=⇒ b = d and there are d n leaves
I Depth-first search for CSPs with single-variable assignments is called backtracking search
I Backtracking search is the basic uninformed algorithm for CSPs
I Can solve n-queens for n ≈ 25

12 / 31
Backtracking search

function Backtracking-Search(csp) returns solution/failure

return Recursive-Backtracking({ }, csp)

function Recursive-Backtracking(assignment, csp) returns soln/failure

if assignment is complete then return assignment
var ← Select-Unassigned-Variable(Variables[csp], assignment, csp)
for each value in Order-Domain-Values(var, assignment, csp) do
if value is consistent with assignment given Constraints[csp] then
add {var = value} to assignment
result ← Recursive-Backtracking(assignment, csp)
if result 6= failure then return result
remove {var = value} from assignment
return failure

13 / 31
Backtracking example

14 / 31
Backtracking example

14 / 31
Backtracking example

14 / 31
Backtracking example

14 / 31
Improving backtracking efficiency

General-purpose methods can give huge gains in speed:

1. Which variable should be assigned next?

2. In what order should its values be tried?
3. Can we detect inevitable failure early?
4. Can we take advantage of problem structure?

15 / 31
Minimum remaining values

Minimum remaining values (MRV):

choose the variable with the fewest legal values

16 / 31
Degree heuristic

Tie-breaker among MRV variables

Degree heuristic:
choose the variable with the most constraints on remaining variables

17 / 31
Least constraining value

Given a variable, choose the least constraining value:

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

Allows 1 value for SA

Allows 0 values for SA

Combining these heuristics makes 1000 queens feasible

18 / 31
Forward checking

Idea: Keep track of remaining legal values for unassigned variables

Terminate search when any variable has no legal values

WA NT Q NSW V SA T

19 / 31
Forward checking

Idea: Keep track of remaining legal values for unassigned variables

Terminate search when any variable has no legal values

WA NT Q NSW V SA T

19 / 31
Forward checking

Idea: Keep track of remaining legal values for unassigned variables

Terminate search when any variable has no legal values

WA NT Q NSW V SA T

19 / 31
Forward checking

Idea: Keep track of remaining legal values for unassigned variables

Terminate search when any variable has no legal values

WA NT Q NSW V SA T

19 / 31
Constraint propagation

Forward checking propagates information from assigned to unassigned variables, but doesn’t
provide early detection for all failures:

WA NT Q NSW V SA T

NT and SA cannot both be blue!

Constraint propagation repeatedly enforces constraints locally

20 / 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

WA NT Q NSW V SA T

21 / 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

WA NT Q NSW V SA T

21 / 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

WA NT Q NSW V SA T

21 / 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

WA NT Q NSW V SA T

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

21 / 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

WA NT Q NSW V SA T

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

I Arc consistency detects failure earlier than forward checking

I Can be run as a preprocessor or after each assignment
21 / 31
Arc consistency algorithm

function AC-3( csp) returns the CSP, possibly with reduced domains
inputs: csp, a binary CSP with variables {X1 , X2 , . . . , Xn }
local variables: queue, a queue of arcs, initially all the arcs in csp

while queue is not empty do

(Xi , Xj ) ← Remove-First(queue)
if Remove-Inconsistent-Values(Xi , Xj ) then
for each Xk in Neighbors[Xi ] do
add (Xk , Xi ) to queue

function Remove-Inconsistent-Values( Xi , Xj ) returns true iff succeeds

removed ← false
for each x in Domain[Xi ] do
if no value y in Domain[Xj ] allows (x,y) to satisfy the constraint Xi ↔ Xj
then delete x from Domain[Xi ]; removed ← true
return removed

O(n2 d 3 ), can be reduced to O(n2 d 2 ) (but detecting all is NP-hard)

22 / 31
Problem structure

NT
Q
WA

SA NSW

V
Victoria

I Tasmania and mainland are independent subproblems

I Identifiable as connected components of constraint graph

23 / 31
Problem structure contd.

I Suppose each subproblem has c variables out of n total

I Worst-case solution cost is n/c · d c , linear in n
I E.g., n = 80, d = 2, c = 20
I 280 = 4 billion years at 10 million nodes/sec
I 4 · 220 = 0.4 seconds at 10 million nodes/sec

24 / 31
Tree-structured CSPs

A E
B D
C F
Theorem
If the constraint graph has no loops, the CSP can be solved in O(n d 2 ) time

I Compare to general CSPs, where worst-case time is O(d n )

I This property also applies to logical and probabilistic reasoning: an important example of the
relation between syntactic restrictions and the complexity of reasoning.

25 / 31
Algorithm for tree-structured CSPs

1. Choose a variable as root, order variables from root to leaves such that every node’s parent
precedes it in the ordering

A E
B D A B C D E F
C F
2. For j from n down to 2, apply RemoveInconsistent(Parent(Xj ), Xj )
3. For j from 1 to n, assign Xj consistently with Parent(Xj )

26 / 31
Nearly tree-structured CSPs
Conditioning: instantiate a variable, prune its neighbors’ domains

NT NT
Q Q
WA WA

SA NSW NSW

V
Victoria V
Victoria

T T

Cutset conditioning: instantiate (in all ways) a set of variables

such that the remaining constraint graph is a tree

Cutset size c =⇒ runtime O(d c · (n − c)d 2 ), very fast for small c

27 / 31
Iterative algorithms for CSPs

Hill-climbing, simulated annealing typically work with

“complete” states, i.e., all variables assigned

To apply to CSPs:

I Allow states with unsatisfied constraints

I Operators reassign variable values
I Variable selection: randomly select any conflicted variable
I Value selection by min-conflicts heuristic: choose value that violates the fewest constraints
I i.e., hillclimb with h(n) = total number of violated constraints

28 / 31
Example: 4-Queens

I States: 4 queens in 4 columns (44 = 256 states)

I Operators: move queen in column
I Goal test: no attacks
I Evaluation: h(n) = number of attacks

29 / 31
Performance of min-conflicts
Given random initial state, can solve n-queens in almost constant time for arbitrary n with high
probability (e.g., n = 10,000,000)

The same appears to be true for any randomly-generated CSP

except in a narrow range of the ratio
number of constraints
R=
number of variables

CPU
time

R
critical
ratio
30 / 31
Summary

I 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
I Backtracking = depth-first search with one variable assigned per node
I Variable ordering and value selection heuristics help significantly
I Forward checking prevents assignments that guarantee later failure
I Constraint propagation (e.g., arc consistency) does additional work to constrain values and
detect inconsistencies
I The CSP representation allows analysis of problem structure
I Tree-structured CSPs can be solved in linear time
I Iterative min-conflicts is usually effective in practice

31 / 31