Can someone refresh me on the concept of fill board?
David Fotland wrote:
I finally found time to try fill_board, and it made the program a little
weaker.
I tested against gnugo level 10, 19x19 board, vs Many Faces with 10K
playouts per move.
Before fill_board it won 1183 of 1411, 84%, 1.9% 95% confidence interval.
After it won 1375 of 1676, 82%, 1.8% 95% confidence interval
This is not enough games to prove that fillboard is weaker, but it is
likely.
My playout move generator implementation:
First, make one of the local tactical suggested moves
If none, make one of the 3x3 local pattern moves
If none, sometimes make a capture (more likely for bigger groups)
If none, fill_board: pick a random empty point. If it is not on the 1st or
2nd line and there are no stones in 8 adjacent points, play it
Otherwise, play a random move (constrained to not fill eyes, etc)
I remember that fill_board in Mogo is important to add variety when there
are many playouts, so perhaps the result would be different with longer time
limits. I still use vanilla UCT with a pretty large exploration constant,
so perhaps UCT is giving me more exploration already than Mogo so I don't
need this.
David
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/