There's currently one technical problem that needs to be solved before a feasible checkers AI can be generated, however: we need to use less space in memory. The Connect 4 AI uses up about 100 MB of memory once it plays up to about 70,000 games; there is simply too much state information to keep it all resident in memory at once. My computer has 96 MB of memory, and crawls to a halt after the memory size blows up.
A side effect of the huge amount of memory taken by the program is that it's also about 100 MB to write out the state information to disk. I plan on using Storable to store the data onto disk, because it should take up less space, and maybe tying the state data to some sort of database such that we don't have to keep it all in memory at once. We could also decide to pack state info into a more concise form than strings; that should save on space, but I doubt that the small savings there (on the order of using roughly 5x less space) would be enough savings to make checkers (or even connect four) viable.
All this space doubles if you choose to use another AI player (as opposed to the Random player) for the second player.
In reply to Re: Re: A simple game AI that learns
by Falkkin
in thread A simple game AI that learns
by Falkkin
| For: | Use: | ||
| & | & | ||
| < | < | ||
| > | > | ||
| [ | [ | ||
| ] | ] |