[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [gnugo-devel] possible reading.c speedups
From: |
Arend Bayer |
Subject: |
Re: [gnugo-devel] possible reading.c speedups |
Date: |
Fri, 8 Feb 2002 10:53:37 +0100 (CET) |
> I made a crude experiment with history heuristic once but without any
> measurable success. Might have been because the implementation was too
> bad.
I think the difficult part will be finding a good heuristic which
successful moves in other variations are still relevant. In chess,
this seems a little easier: the mere fact that the move "From X to Y"
is still possible should already be a good hint that it is also relevant.
(Chess engines store the history table as a 64x64 array.) An idea might be
to make the history table as an array history[max_depth][BOARDMAX] and only
take into account moves that were successful at a similar depth. Hmm.
> > Some of these techniques should be applicable to owl as well, where
> > they might enable us to increase the maximal number of moves tried.
>
> I don't think you mean what you're saying here. More sophisticated
> move selection should reduce the *number of moves* that can be tried
> within a given amount of time, since each move takes more time to
> generate. What it increases is rather the size of the tree which is
> analyzed.
>
What I meant to say is: Maybe we could increase accuracy of owl by
changing MAX_MOVES from 3 to 4 (or increase the depth where branching
still happens etc.), without loosing speed or getting more owl node limit
cut offs compared to today.
Arend