|
From: | Gunnar Farnebäck |
Subject: | Re: [gnugo-devel] on the road to 3.8 |
Date: | Tue, 23 Oct 2007 23:07:33 +0200 |
User-agent: | Mozilla-Thunderbird 2.0.0.6 (X11/20071008) |
Joseph Piche wrote:
With regard to bug #175, I have to agree that G5 on move 52 is awkward, and that a more appropriate move would be B8 or F1. However, after viewing output with -t -t, it looks like gnugo is choosing G5 to increase territory, even though Chinese rules are being used. Gnugo should see the territories at B8 and F1 as needing to be strengthened--which with the F1 territory, it does at endgame (which might actually be unnecessary.
F1 is safe and doesn't need defending, it's only strengthened as part of the --capture-all-dead process. B8 on the other hand seems necessary to avoid seki and that corner position should go into the seki test suite, cf. e.g. seki:901-913.
I do admit though that I tested it under 3.7.10, not CVS. I haven't worked with CVS much, so could someone post the command I need to use to grab the source.
For instructions, see https://savannah.gnu.org/cvs/?group=gnugo
With bug #160, I have run into this too at times. One possible (yet not very good) options is to add a flag and (maybe) gtp command for an individual move time limit, like --time-limit X or time_limit X, where if gnugo reached that limit when trying to decide a move, it would simply choose one sort of at random.
This is much harder than it sounds. GNU Go just doesn't have any infrastructure for aborting a search (or even the overall move generation) based on time limits or other external factors.
/Gunnar
[Prev in Thread] | Current Thread | [Next in Thread] |