Final year project

This report looks at how Fini te State Methods solve Constraint Sat isfact ion Problems, using the Xerox Fini te State Toolki t and a comparable open source f in i te state toolk i t developed by Mans Hulden, foma. The report then compares the performance with tradit ional Constraint Sat isfact ion Problem Methods, such as a pr imit ive Backtracking method, and a more advanced Part ia l Instant iat ion method. ! Specif ical ly, th is report looks at i f F in i te State Methods are a v iable al ternat ive for solv ing Constraint Sat isfact ion Problems and i f not, why not. This is done by using the Zebra Puzzle, and a smal ler version, the Mini-Zebra Puzzle, as example puzzles. ! This report found that Tradit ional methods wi l l perform wel l in a solut ion-r ich environment, whereas Fini te State Methods wi l l perform wel l regardless of number of solut ions. This tradeoff is best made on a per-problem basis, on the assumption that intr insic knowledge is gained a pr ior i about a rough number of solut ions that exist to the problem. In the case that no pr ior knowledge exists, i t then depends on the number of solut ions required. Fini te State Methods perform best when al l solut ions are desired, and Tradit ional Methods perform wel l when merely the f i rst solut ion is required. ! This report a lso found that Fini te State Methods could be great ly opt imised by developing entai lments to see which constraints are necessary to gain a solut ion, before commencement of evaluat ion. !