[Top][All Lists]

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

Re: [gnugo-devel] Slow...

From: Daniel Bump
Subject: Re: [gnugo-devel] Slow...
Date: Thu, 1 Nov 2001 12:41:51 -0800

> The latest CVS version is quite slow.  Here's the slowest problems
> from neurogo.tst:

Neurogo was a notoriously slow game. Around moves 102 and 104 the
situation in the upper left is very complicated. During the
match I think we started wondering whether GNU was ever going 
to move.

Yes, we are getting slower. This happened on the way to 3.0.
We introduced the owl code (known then as mouse) and it
looked very promising but also very slow. As we tuned it the
program got stronger and slower until it was so slow we
couldn't release our program. Then we started looking for
speedups. Notably the persistent caching. Eventually we got
a version we were happy with but it took us over a year.

> Are there any quantitative performance goals?

Not per se. Enhancing strength and speed are often competing
goals. In the early stage of the release cycle, enhancing
strength is more important. Near release time, speedups are
the more important.

It's not imperative that level 10 be of acceptable speed
to a human opponent. But it would be nice if this were
true. Currently level 7 is fast enough for opponents who
expect snappy responses. But it's substantially weaker.
I think when we see a bad game on NNGS it is often true
that some of the moves are played at a lower level.

We asked David Fotland how fast he thought humans expected
their computers to play, and I think he felt that it's OK if
in the complicated middlegame you have some moves of the
order 30 seconds or more.


reply via email to

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