A feature request related to this:  In uct_evaluate, don't process moves
that are symmetrically-equivalent to moves that have already been processed.


On Thu, Jul 14, 2011 at 9:25 PM, Michael Williams <
[email protected]> wrote:

> I noticed some unusual beharior from Pachi on 6x6 with komi 4.  I
> understand that this is not a format that you are targeting or have tested
> extensively, but I thought I would mention it in case it is symptomatic of
> something that may also come into play at other sizes.
>
> I set up the board, played the natural first move at D4, and then called
> uct_evaluate:
>
>
> > ./pachi -d 0
> boardsize 6
> =
>
> komi 4
> =
>
> clear_board
> =
>
> play b d4
> =
>
> uct_evaluate white
> = A1 0.094
> B1 0.156
> C1 0.165
> D1 0.194
> E1 0.143
> F1 0.125
> A2 0.150
> B2 0.204
> C2 0.262
> D2 0.257
> E2 0.223
> F2 0.154
> A3 0.155
> B3 0.261
> C3 0.632
> D3 0.256
> E3 0.254
> F3 0.154
> A4 0.213
> B4 0.252
> C4 0.230
> F6 0.095
> E4 0.277
> F4 0.188
> A5 0.142
> B5 0.242
> C5 0.235
> D5 0.247
> E5 0.208
> F5 0.162
> A6 0.116
> B6 0.135
> C6 0.150
> D6 0.226
> E6 0.152
>
>
>  Everything looks good.  The highest-rated reply is C3.  But when I call
> genmove:
>
>
> genmove w
> = D5
>
>
> It never returns C3.  It returns either D5 or E5.  The same problem occurs
> with komi 3.5.
>
>
>
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to