Wednesday, January 25, 2012

1-ply benchmark for Player 2.1

I played 0- and 1-ply Player 2.1 in single-game match mode (where gammons & backgammons count as one point) against pubEval to see how it compares to GnuBG.

For 1k games, 0-ply Player 2.1 wins 68.8% of games, and 1-ply Player 2.1 wins 70.2% of games. So adding the 1-ply lookahead improves the win percentage by 1.4%, so a bit less improvement than I expected from the equivalent test for GnuBG. (And I note my 1-ply player doesn't even get to the performance of 0-ply GnuBG (71.9%).)

I also think my lookahead algorithm is inefficient now. It's really really slow for 2-ply - so slow I can't plausibly calculate statistics. Needs some trimming!

No comments:

Post a Comment