Re: [gnugo-devel] Parallelizing GNU Go

[bump]

> > Is there a planned time frame for 3.4?  Or specific goals?  I don't have
> > any particular opinions on the matter; just curious as to what the plan
> > is.
> 
> In my opinion the main showstopper is the semeai code. If we could get
> the reliability of that up to par with the rest of engine I would be
> content to start the release polishing (stress testing, docs
> revisions, etc.).

We were hoping to release this spring but (speaking for myself)
I became dissatisfied with the way GNU Go plays sometimes. In
the past each release has been at least a 2 stone improvement
and I think the improvement over 3.2 may be about 1.5 stones
at this point.

The first example that comes to mind of this dissatisfaction
that I mentioned is the game uhu-gnugo-3.3.16-200302112129.sgf
where the main problems are semeai related. So I agree with
Gunnar about semeai.

I don't think we can make the results of owl_analyze_semeai() as
reliable as owl_attack() and owl_defend() before 3.4 but we can
try. Also there may be ways we can mitigate its deficiencies. 
For example, if the opponent attacks our semeai dragon to treat
the opponent's semeai dragon as thrashing.

Dan