[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] Re: computer-go: Computer Go Ladder
From: |
Arend Bayer |
Subject: |
Re: [gnugo-devel] Re: computer-go: Computer Go Ladder |
Date: |
Wed, 16 Apr 2003 11:03:48 -0400 (EDT) |
Evan wrote:
> On the more general subject of tournament play, I would like to propose a
> --tournament option. This would have (potentially) a number of effects
> that attempt to get the strongest engine we can. For example, we could
> add a pattern classification to joseki patterns such that some josekis are
> not played during tournaments. For example, we could reduce the
> probability of the high approach to hoshi, not play the avalanche, only
> play one simple response to the avalanche (the solid connection), etc. I
> think having a varied engine that plays a larger variety of plays is
> interesting in general, and allows us to find a wider variety of problems,
> but for tournament play we should probably play the strongest moves we can
> find, and avoid positions that may be difficult for gnugo.
Don't forget that indeterminacy is also important for tournament play, this
has been discussed on this list a couple of times, I believe.
What might be interesting is a compile time tournament option that turns
off all asserts, tracing and sgf tracing. Would certainly give a measurable
speed up.
(This is all far less important for tournaments than a good level adjustment
scheme that would reliably allow GNU Go to almost fully use the allotted
time in tournament matches, however.)
Arend