[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [gnugo-devel] Metamachine's first game

From: bump
Subject: Re: [gnugo-devel] Metamachine's first game
Date: Wed, 13 Nov 2002 08:45:10 -0800

I have put up oracle_3_12.1e, which is a bug fix revision of
oracle_3_12.1d. As I mentioned in my previous message,
oracle_3_12.1d would crash if you used it to load and
analyze an sgf file.

Now you can run:

gnugo -l ManyFaces3-gnugo-3.3.11-200211071935.sgf -L25
    --quiet --metamachine --limit-search D14 -t -o vars.sgf

and notice that the metamachine makes the right move at
C14. (If you omit the search limitation, it still moves in
the same area but plays the wrong move at B10.)

> Haven't paid close attention to how the metamachine works.
> Does the main process think in parallel to the metamachine,
> or does gnugo ask the metamachine a question then wait for
> an answer ?

In the current implementation, one process thinks, the other
just asks questions and does some minmax search. It waits
for an answer. For what it's doing now, it would be possible
and probably not difficult to remove the oracle from this
setup. Everything would be done by a single process.

Other applications might not have this feature. For example
in the pattern database application, which was the first one
I posted about, both engines think, though not
simultaneously.  A trymove in one engine is paralleled by
actually playing the move in the other one. The function
of the oracle is just to make the owl code available at
deeper levels of reading. 

I'd like to see a real multitasking setup where different
processes could be thinking at the same time on different

Short of that, the immediate direction I have in mind for
these experiments is for the controlling engine to maintain
local move trees.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]