sep 24
discuss projects
- due in one week, with presentations
- Isaac - gameworld generation via simulated annealing, e.g. require that the generated world maze be connected
- Ryon - generation of "what do I want" tree based on previous choices, searching for best next choice (python)
- Patrick - bidirectional iterative deepening search on connecting web sites via links (python)
- Alec - genetic algorithm for crypto with "time to solve" as a goodness of fit (lisp)
- Ben - something to do with searching in game of "set" space (perl)
- Liz - genetic algorithm
- Richard - tic tac toe comparison of search strategies (lisp)
- Abe - also likes bidirectional web search (lisp)
- Carrie - ?
- Jim - perhaps constraint satisfaction ?
4.14
chap 5 - constraint satisfaction problems
- Variables X_i; 1 < i < N
- Constraints C_j; 1 < j < M
- State is assignment X_1=foo X_2=bar, .
- Solution is an assignment that satistifies all constraints; may also require maximizes a given function of X's.
- Can write some standard algorithms to do these well, and then plug in a particular problem.
- Branches of tree assign one variable (since order doesn't matter)
- Tree is N deep.
- Search is generically dept-first backtracking.
- Refinements:
- MRV - minimum remaining values (pick variable with least choices first)
- forward checking - remove inconsistent values from unassigned variables
- constraint propogation and arc consistency
- special constraints: alldiff, atmost
- conflict-directed backjumping
- Local search jumps around between bottom nodes of the tree.
- automatic searching for tree decomposition
(SOLVE (MAKE-NQUEENS-PROBLEM) 'CSP-BACKTRACKING-SEARCH)
chap 6 - adversarial search
(RUN-GAME (MAKE-TTT_GAME))
sep 26
