[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [computer-go] random evaluation results
Here are some different results. The only difference is that the
program calls the (random) evaluator after 2 passes inside the search.
I still find it amazing that random evaluation performs so well with
extra depth even though I knew of this finding before. I just wasn't
sure how much of an effect it would have in GO.
Will test with bigger boards next. Will it be better, worse or about
the same? Any predictions?
Would anyone be willing to verify my results using the same
parameters? Gunnar is doing something similar, but he is using a
whole different methodology which would take me some time to
duplicate.
- Don
r2 51.700 1000 games
r3 71.300 1000 games
r4 73.400 1000 games
r5 81.000 1000 games
r6 91.000 1000 games
r7 85.841 113 games
> I have no clue about what's going on at depth 7. I cannot find a bug
> that would explain this.
But I think it has something to do with not ending the game INSIDE the
search after N passes ...
I ran several games with random evaluation and have some results to
report.
boardsize: 5
komi: 0.5
ko: simple
game over: after 2 passes (but not inside search)
eval: #define EVAL (8192 - (rand() & 0x3fff))
scoring: chinese (tromp/taylor) - no dead stone resolving.
r2 44.300 1000 games
r3 62.200 1000 games
r4 68.800 1000 games
r5 73.000 1000 games
r6 92.200 1000 games
r7 38.616 549 games
r2 is a 2 ply search, r3 is 3 ply etc. All players are playing
against r1 which is a 1 ply search which is the equivalent of simply
playing a random legal move.
I have no clue about what's going on at depth 7. I cannot find a bug
that would explain this.
- Don
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/