1. Constraint Satisfaction Basics strongly influenced by Dr. Rina Dechter: “Constraint Processing”

2. complete search space of a problem  variables V = {v 1, v 2, …., v n }  domains D = {D 1, D 2, …., D n }, v i  D i  search space T = D 1 X D 2 X…. X D n  size of search space |D 1 |. |D 2 |. ….. |D n | if there are no constraints, any solution in T is feasible

3. constraint satisfaction problems  variables V = {v 1, v 2, …., v n }  domains D = {D 1, D 2, …., D n }, v i  D i  constraints C = {C 1, C 2, …., C k } C i is a relation on scope S i  V C i puts constraints on some variables in the problem  search space T = D 1 X D 2 X…. X D n a solution in the search space T whose values violate a constraint is infeasible a constraint satisfaction problem is often called a constraint network

4. example: 4 queens problem column version  variables V = {v 1, v 2, v 3, v 4 }  domains D 1 = D 2 = D 3 = D 4 = {1,2,3,4}  constraints: “no two queens should attack one another” C = {C 1, C 2, …., C 6 } C 1 is a relation on scope S 1 = {v 1, v 2 } (constraint between first and second queens) enumerated: {(1,3),(1,4),(2,4),(3,1),(4,1),(4,2)} algebraic: { (v 1,v 2 ) | v 1 ≠ v 2, v 1 ≠ v 2 ± 1 }  others defined likewise v 1 v 2 v 3 v 4 12341234

5. describing constraints  scope S i :the set of variables on which a constraint C i is defined  scheme S = {S 1, S 2, …, S k } set of all scopes on which constraints are defined  arity of a constraint C i is the size of its scope |S i | unary constraint on one variable binary constraint on two variables (4 queens eg) n-ary constraint on n variables* *n-ary constraints can be rewritten as (many) binaries

6. simple scheduling problem five tasks to schedule, T1, T2, T3, T4, T5  each lasts one hour  each may start at 1PM, 2PM, 3PM  tasks can be executed simultaneously except: T1 starts after T3 T3 starts before T4 and after T5 T2 cannot be concurrent with T1 or T4 T4 cannot start at 2PM

7. simple scheduling problem five tasks to schedule, T1, T2, T3, T4, T5  variables? domains?  constraints? scopes? arity?

8. constraint graphs  vertices: variables  edges: (binary) variable scopes 4 queensscheduling problem v1v1 v1v1 v2v2 v2v2 v4v4 v4v4 v3v3 v3v3 T1 T2 T3 T4 T5

9. crossword puzzle (after Dechter) HOSESLASERSHEETSNAIL STEERALSOEARNHIKE IRONSAMEEATLET RUNSUNTENYES MEITNOUS 1 1 2 2 3 3 4 4 5 5 7 7 6 6 8 8 9 9 12 10 11 13

10. crossword puzzle HOSESLASERSHEETSNAIL STEERALSOEARNHIKE IRONSAMEEATLET RUNSUNTENYES MEITNOUS 1 1 2 2 3 3 4 4 5 5 7 7 6 6 8 8 9 9 12 10 11 13 variables: 13 (letters) domains: alphabet constraints: S 1 {1,2,3,4,5} C1 {(H,O,S,E,S), (L,A,S,E,R), (S,H,E,E,T), (S,N,A,I,L), (S,T,E,E,R)}

11. crossword puzzle HOSESLASERSHEETSNAIL STEERALSOEARNHIKE IRONSAMEEATLET RUNSUNTENYES MEITNOUS 1 1 2 2 3 3 4 4 5 5 7 7 6 6 8 8 9 9 12 10 11 13 S 1 {1,2,3,4,5} arity 5 S 2 {3,6,9,12} S 3 {5,7,11} S 4 {8,9,10,11} S 5 {10,13} S 6 {12,13}

12. crossword puzzle HOSESLASERSHEETSNAIL STEERALSOEARNHIKE IRONSAMEEATLET RUNSUNTENYES MEITNOUS 1 1 2 2 3 3 4 4 5 5 7 7 6 6 8 8 9 9 12 10 11 13 partial solution satisfying C 4 and C 5 over S 4  S 5 {8,9,10,11,13} {(S,A,M,E,E)}

13. graphs for arity > 2  hypergraph multiple nodes per “hyperedge” 1 1 2 2 3 3 4 4 5 5 7 7 6 6 8 8 9 9 12 10 11 13

14. graphs for arity > 2  dual of hypergraph nodes are constraints edges are common variables 1 1 2 2 3 3 4 4 5 5 7 7 6 6 8 8 9 9 12 10 11 13 1,2,3,4,5 5,7,11 3,6,9,12 8,9,10,11 12,13 10,13 3 11 12 10 59 13

15. crossword puzzle another formulation HOSESLASERSHEETSNAIL STEERALSOEARNHIKE IRONSAMEEATLET RUNSUNTENYES MEITNOUS 1 1 2 2 3 3 4 4 5 5 7 7 6 6 8 8 9 9 12 10 11 13 variables: 6 domains: words by length constraints: crossings S 1 {1,3}S 2 {1,5}S 3 {10,12} S 4 {3,8}S 5 {3,12}S 6 {5,8} S 7 {10,8} all binary constraints

16. binary constraint networks only unary and binary constraints  constraint deduction inferring new constraints from initial set 1.constraints between unconstrained variables 2.tightening of existing constraints

17. constraint deduction example: V = { v 1,v 2,v 3 } D 1 = D 2 = D 3 = { red, green} C 1 : {(v 2,v 1 )|v 2 ≠v 1 } = {(red, green),(green, red)} C 2 : {(v 1,v 3 )|v 1 ≠v 3 } = {(red, green),(green, red)} solutions: {(red, green, green), (green, red, red)} red green red green red green v1v1 v2v2 v3v3

18. constraint deduction example: new constraint network with same solutions --> better for partial solutions (more later) red green red green red green v1v1 v2v2 v3v3 red green red green red green v1v1 v2v2 v3v3 inferred constraint: v 2 = v 3

19. constraint composition given two binary* constraints C 1, C 2 on scopes S 1 = {x,y} and S 2 = {y,z} then the composition C 3 =C 1. C 2 is defined on S 3 = {x,z} C3 = {(a,b)| a  D x, b  D z,  c  Dy such that (a,c)  C 1 and (c,b)  C 2 } e.g.,C 1 = {(red, green),(green, red)} C 2 = {(red, green),(green, red)} C 3 =C 1. C 2 = {(red, red), (green, green)} *also works for a unary and a binary

20. inferring with constraints who owns the zebra? variables domains constraints constraint graph