The C reference bot indeed had a simple bug. Everything was fine except for when the final move choice was made, I forgot to check for eye.
I am surprised it played as strongly as it did because it seems like it would have always filled eyes, but the logic of not selecting a move that has no playouts associated with it kicked in. So it ended up being roughly only 15 ELO weaker. That explains why the other statistics seemed to match perfectly, there was nothing wrong with the playouts. It was outside 4 standard deviations after many thousand games. Now it is within 2 SD after about 8000 games. I'm going to let it run for a few more thousand games and see if this continues to hold up because the javabot is scoring 51.01 percent, just barely within 2 standard deviations. - Don
signature.asc
Description: This is a digitally signed message part
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
