[Top][All Lists]

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

Re: [gnugo-devel] questions regarding cosmic option in gnugo 3.4

From: bump
Subject: Re: [gnugo-devel] questions regarding cosmic option in gnugo 3.4
Date: Sun, 2 May 2004 21:13:09 -0700

> I didn't find much explanation on the cosmic option in the documentation.
> So I have some questions.
> gnugo 3.4 do not use cosmic as default, in order to enable it, I need to
> set --enable-cosmic-gnugo at ./configure, correct?


> I noticed cosmic was a default in previous version of gnugo, why is it not
> a default in 3.4?

It was experimental. We made it the default for a while to
try it out. This was during September 4, 2003--November 5, 2003
while gnugo-3.5.1 was the current version. Note how the rating
graph dipped during 3.5.1 and started to recover as soon as
3.5.2 was released.

I think it is safe to say that the cosmic option makes GNU
Go weaker. 

When the cosmic option is enabled, GNU tries to build
moyos in the center. The term ``cosmic Go'' refers to
Takemiya style. Takemiya had a lot of success with
this style but GNU Go isn't strong enough to play
it. It concedes a lot of territory to build a moyo
that is full of holes.

> most importantly, I am puzzled by the variable cosmic_importance in
> engine/influence.c file.

One thing that may be confusing is that only every other
move is ``cosmic.'' In effect there are two modes to the
engine which alternate.

The cosmic option has good success in gnugo-gnugo games.
But against other opponents, it does not work well.

The configure option --enable-large-scale (also
by Stephane Nicolet) is similarly experimental but I think
much better, in fact I think it is safe to say that it
makes the engine stronger, though it may be too expensive
in terms of speed.

BTW the current CVS fixes a bug in 3.5.6 which causes
3.5.6 to generate about 5% more owl nodes than it should.
This will be fixed in 3.5.7 but meanwhile I suggest using
the current CVS version instead of 3.5.6.


reply via email to

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