|
|
Post by Steve Draper on Jul 12, 2015 8:29:18 GMT -8
|
|
|
|
Post by alandau on Jul 12, 2015 14:53:00 GMT -8
Nice work! I tried something like this in Alloy a while back but it wasn't nearly as polished; I think my approach to integrating the results back into MCTS wasn't as effective. (In particular, I was focusing on a type of search where the opponent always made an irrelevant play/tenuki, and then prioritized the search towards those play sequences.) There wasn't a locality constraint, so the initial search didn't get as far, either; though I did try a "novelty"-based approach to building out longer move sequences, which might be another way to apply this. (The simple version is to try moves that weren't available during your last choice of moves.)
|
|
