For my optimization class last semester, I built an entire chess game using HTML and JavaScript. I also wrote an artificial intelligence algorithm to play as a computer player. My algorithm only does a two-ply (easy setting) or a three-ply (hard setting) search. It has about 20 different parameters it uses to score each move and decide which move it should take.
For my optimization class project, I also created a real-value genetic algorithm to optimize these 20 parameters and find the "best" settings for the computer player.
With more computing power, the computer player could be increased to a three- or four-ply search. But even with just the two-ply search I ran my genetic algorithm on, it took a pretty significant amount of computing time! We had 20-30 cores cranking away for about 2 weeks. Here are the stats after completing the genetic algorithm:
| Total Generations: | 31 |
| Total Players: | 128 |
| Games Played: | 1,832 |
| Turns Taken: | 637,900 |
| Processing Time: | 3,830 core-hours (160 continuous days) |
There are still a few small bugs that need to worked out in the game (castling sometimes doesn't work properly, en passant isn't possible, and a stalemate is treated as a checkmate). But overall, it's been a fun project to work on!
(Of course, this game is fastest in Chrome, and awfully slow in Internet Explorer.
No comments:
Post a Comment