Re: [gnugo-devel] Playing with cosmic gnugo

[Evan Berggren Daniel]

On Sun, 10 Aug 2003, SP LEE wrote:

> While it's easy for me to win GG with 9 stone komi before cosmic patch, I
> find it's very difficult for me to win cosmic gnugo with 9 stone komi. I
> wonder how others feel about cosmic gnugo and I'm really interested what the
> results will be on nngs games after the cosmic option is used on nngs.

The reports I hear about the cosmic patch are very interesting; I'll
probably give it a try sometime soon.

However, I have two major reservations about the patch:

First, I don't like the idea of having a move number dependency, both in
general and especially in the case of odd / even move numbers.  For things
that decline over time, I think that is a shortcut to understanding how
much the game has progressed; I believe that a complex joseki in one
corner that left the board open could take as many moves as a realtively
relaxed fuseki that settled much of the board; clearly the two games are
in different states, and the reverse is also true.  I think move number is
an acceptable way to handle the issue, so long as there are long term
plans to do something more sophisticated.  Is there any way to remove the
odd / even differences?  I think that gnugo should play without regard to
that, if only to make regressions more predictable.

My second worry about the patch is regression changes; unless I'm missing
something, the regression change is very small for the patch, yet people
are saying there is a substantial strength increase.  I would want to see
some regression tests that the cosmic patch both passes and fails added to
the test suite, if only to give a better sense of exactly what sort of
moves are being improved or worsened.  These should be fairly simple to
create by replaying some nngs games with the cosmic patch, or replaying
some cosmic games without the patch.

Anyway, that's just my .02, and I'm not really the one you need to
convince anyway ;)

Evan Daniel