1. CSC 8520 Spring 2013. Paula Matuszek CS 8520: Artificial Intelligence Search 3: Constraint Satisfaction Problems Paula Matuszek Spring, 2013

2. 2 CSC 8520 Spring 2013. Paula Matuszek Outline Constraint Satisfaction Problems (CSP) Backtracking search for CSPs Local search for CSPs

3. 3 CSC 8520 Spring 2013. Paula Matuszek Constraint satisfaction problems (CSPs) Standard search problem: –state is a "black box“ – any data structure that supports successor function, heuristic function, and goal test CSP: –state is defined by variables X i with some values from domain D i –a set of constraints C i specifies allowable combinations of values for subsets of variables –a consistent state violates none of the constraints C –a complete assignment has values assigned to all variables. –A solution is a complete, consistent assignment. Simple example of a formal representation language Allows useful general-purpose algorithms with more power than standard search algorithms

4. 4 CSC 8520 Spring 2013. Paula Matuszek Example: Map-Coloring Variables WA, NT, Q, NSW, V, SA, T Domains D i = {red,green,blue} Constraints: adjacent regions must have different colors e.g., WA ≠ NT, or (WA,NT) in {(red,green),(red,blue),(green,red), (green,blue),(blue,red),(blue,green)}

5. 5 CSC 8520 Spring 2013. Paula Matuszek Example: Map-Coloring Solutions are complete and consistent assignments, e.g., WA = red, NT = green,Q = red,NSW = green,V = red,SA = blue,T = green

6. 6 CSC 8520 Spring 2013. Paula Matuszek Constraint graph Binary CSP: each constraint relates two variables Constraint graph: nodes are variables, arcs are constraints

7. 7 CSC 8520 Spring 2013. Paula Matuszek Varieties of CSPs Discrete variables –finite domains: n variables, domain size d  O(d n ) complete assignments e.g., Boolean CSPs, incl.~Boolean satisfiability (NP-complete) –infinite domains: integers, strings, etc. e.g., job scheduling, variables are start/end days for each job need a constraint language, e.g., StartJob 1 + 5 ≤ StartJob 3 Continuous variables –e.g., start/end times for Hubble Space Telescope observations –linear constraints solvable in polynomial time by linear programming algorithms from operations research

8. 8 CSC 8520 Spring 2013. Paula Matuszek 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., cryptarithmetic column constraints Global constraints: arbitrary # of constraints, not necessarily all the variables in a problem –e.g., AllDiff: all values must be different. Sudoku rows, cols, squares

9. 9 CSC 8520 Spring 2013. Paula Matuszek Example: Cryptarithmetic Variables: F T U W R O X 1 X 2 X 3 Domains: {0,1,2,3,4,5,6,7,8,9} Constraints: Alldiff (F,T,U,W,R,O) –O + O = R + 10 · X 1 –X 1 + W + W = U + 10 · X 2 –X 2 + T + T = O + 10 · X 3 –X 3 = F, T ≠ 0, F ≠ 0

10. 10 CSC 8520 Spring 2013. Paula Matuszek Real-world CSPs Common problems: –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 May also include preference constraints: constraint optimization

11. 11 CSC 8520 Spring 2013. Paula Matuszek Standard search formulation (incremental) Let's start with the straightforward approach, then fix it: States are defined by the values assigned so far Initial state: the empty assignment { } Successor function: assign a value to an unassigned variable that does not conflict with current assignment  fail if no legal assignments 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 - l )d at depth l, hence n! · d n leaves

12. 12 CSC 8520 Spring 2013. Paula Matuszek Backtracking search Variable assignments are commutative}, i.e., [ WA = red then NT = green ] same as [ NT = green then WA = red ] Only need to consider assignments to a single variable at each node  b = d and there are d n leaves Depth-first search for CSPs with single-variable assignments is called backtracking search Backtracking search is the basic uninformed algorithm for CSPs Can solve n-queens for n ≈ 25

13. 13 CSC 8520 Spring 2013. Paula Matuszek Backtracking search

14. 14 CSC 8520 Spring 2013. Paula Matuszek Backtracking example

15. 15 CSC 8520 Spring 2013. Paula Matuszek Backtracking example

16. 16 CSC 8520 Spring 2013. Paula Matuszek Backtracking example

17. 17 CSC 8520 Spring 2013. Paula Matuszek Backtracking example

18. 18 CSC 8520 Spring 2013. Paula Matuszek 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?

19. 19 CSC 8520 Spring 2013. Paula Matuszek Most constrained variable Most constrained variable: choose the variable with the fewest legal values a.k.a. minimum remaining values (MRV) heuristic

20. 20 CSC 8520 Spring 2013. Paula Matuszek Most constraining variable Tie-breaker among most constrained variables Most constraining variable: –choose the variable with the most constraints on remaining variables

21. 21 CSC 8520 Spring 2013. Paula Matuszek Least constraining value Given a variable, choose the least constraining value: –the one that rules out the fewest values in the remaining variables Combining these heuristics makes 1000 queens feasible

22. 22 CSC 8520 Spring 2013. Paula Matuszek Search Plus Inference Even with these heuristics, a straight backtracking search isn’t very efficient Can improve performance by doing some reasoning or inference. Typically, constraint propagation. Constraints restrict the possible values for assignment, which may in turn restrict the possible values for other assignments. Local consistency Limits search space; may actually give solution.

23. 23 CSC 8520 Spring 2013. Paula Matuszek Forward checking Idea: –Keep track of remaining legal values for unassigned variables –Terminate search when any variable has no legal values

24. 24 CSC 8520 Spring 2013. Paula Matuszek Forward checking Idea: –Keep track of remaining legal values for unassigned variables –Terminate search when any variable has no legal values

25. 25 CSC 8520 Spring 2013. Paula Matuszek Forward checking Idea: –Keep track of remaining legal values for unassigned variables –Terminate search when any variable has no legal values

26. 26 CSC 8520 Spring 2013. Paula Matuszek Forward checking Idea: –Keep track of remaining legal values for unassigned variables –Terminate search when any variable has no legal values

27. 27 CSC 8520 Spring 2013. Paula Matuszek Constraint propagation Forward checking propagates information from assigned to unassigned variables, but doesn't provide early detection for all failures: NT and SA cannot both be blue! Constraint propagation repeatedly enforces constraints locally

28. 28 CSC 8520 Spring 2013. Paula Matuszek 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

29. 29 CSC 8520 Spring 2013. Paula Matuszek 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

30. 30 CSC 8520 Spring 2013. Paula Matuszek Arc consistency If X loses a value, neighbors of X need to be rechecked Simplest form of propagation makes each arc consistent X  Y is consistent iff for every value x of X there is some allowed y

31. 31 CSC 8520 Spring 2013. Paula Matuszek Arc consistency 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 Simplest form of propagation makes each arc consistent X  Y is consistent iff for every value x of X there is some allowed y

32. 32 CSC 8520 Spring 2013. Paula Matuszek Arc consistency algorithm AC-3 Time complexity: O(n 2 d 3 )

33. 33 CSC 8520 Spring 2013. Paula Matuszek Path Consistency Consider coloring Australia with 2 colors. –Clearly not possible. WA, for example, touches two other countries. –AC-3 will not detect this immediately. Path consistency extends it to look at triples of variables. A two-variable set X, Y is path-consistent with a third variable Z iff –for every value a of X and b of Y which satisfies the constraints on {X, Y} –there is a value c of Z which satisfies the constraints on {X,Z} and {Z,Y}. Can look at it as looking at a path from X to Y with Z in the middle. Path consistency algorithm PC-2 is extension of AC-3.

34. 34 CSC 8520 Spring 2013. Paula Matuszek K-Consistency Can extend this to an arbitrary number k of variables a CSP is k-consistent if for any set of k - 1 variables and a consistent assignment to those variables, a consistent value can be assigned to the kth variable. However, costly. In practice, don’t usually go beyond AC-3 (2-consistency) and PC-2 (3-consistency)

35. 35 CSC 8520 Spring 2013. Paula Matuszek Global Constraints Some global constraints have specialized algorithms or heuristics –AllDiff: all variables are distinct Sudoku rows, columns, boxes –AtMost: resources are constrained Scheduling 10 people on 4 tasks

36. 36 CSC 8520 Spring 2013. Paula Matuszek AllDiff If there are m variables and n possible values, any time m>n it is inconsistent Algorithm: –find all variables with singleton domains (there is only one value which satisfies the constraint) –remove that value from domains of remaining variables If m>n or any domain is empty, inconsistent. Classic way to solve easy Sudoku puzzles Reduces search space for more difficult puzzles.

37. 37 CSC 8520 Spring 2013. Paula Matuszek Resource Constraints Constraints expressed as maximum a sum of values can reach: 10 people assigned to 4 tasks Each task can have 1,2,3...10 people initially. –T1 = [1, 2,...10]; T2 = [1,2,...10] etc. Sum minimum in allowable domain for each task; if >10, inconsistent Propagate by removing any value in a domain which is inconsistent with minimum value in other domains –T1 = [4..10], etc. 37

38. 38 CSC 8520 Spring 2013. Paula Matuszek Bounds Propagation For larger problems: 420 people on planes with capacities 165 and 385 passengers. Individual variables get min and max instead of enumerated values –Plane1 = [0, 165]; Plane2 = [0, 385] Propagating the bounds gives us –Plane1 = [35, 165]; Plane2 = [255, 385] 38

39. 39 CSC 8520 Spring 2013. Paula Matuszek Local search for CSPs Hill-climbing, simulated annealing typically work with "complete" states, i.e., all variables assigned 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

40. 40 CSC 8520 Spring 2013. Paula Matuszek Example: 4-Queens States: 4 queens in 4 columns (4 4 = 256 states) Actions: move queen in column Goal test: no attacks Evaluation: h(n) = number of attacks Given random initial state, can solve n-queens in almost constant time for arbitrary n with high probability (e.g., n = 10,000,000)

41. 41 CSC 8520 Spring 2013. Paula Matuszek Local Search Works best when solutions are dense –8 queens has many solutions; good –Proper Sudoku puzzles have one; bad All the usual problems, and solutions, for local search Constraint weighing adjusts weight of constraints which have not been solved: help focus on harder problems. Especially useful when problem gets small changes –airline scheduling when PHL closes

42. 42 CSC 8520 Spring 2013. Paula Matuszek 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 often effective in practice