[Computer-go] Congratulations to Zen!
terrymcintyre at yahoo.com
Mon Apr 4 10:08:08 PDT 2011
A lot of interesting questions. It seems like the best way to resolve them is to
work on building an extensive book?
Is there a way to parameterize the book, to say "given a komi of 7, the book
recommends this, but for komi of 7.5, this line is preferred . . ."?
Terry McIntyre <terrymcintyre at yahoo.com>
Unix/Linux Systems Administration
Taking time to do it right saves having to do it twice.
From: "valkyria at phmp.se" <valkyria at phmp.se>
To: computer-go at dvandva.org
Sent: Mon, April 4, 2011 12:10:07 PM
Subject: Re: [Computer-go] Congratulations to Zen!
Valkyria was even more extreme losing 8 times as black and only 2 times as
I think it has to do with how the opening book is made for komi 7.5 where white
clearly is favored.
In order to not get killed by good books (such as Fuego) black as the underdog
needs to play surprising moves that leads to a long good fight where hopefully
the other programs has more bugs and biases than the own program. It could be
that i have overdone this for Valkyria. That is it does play not strong enough
moves early in the games.
For white it is enough to find one good strong answer to every position that are
likely to appear, so it is much easier to prepare. The choice to use several
moves for black is very hard to do.
Also building a book for 7.5 means that "7.0 jigo lines" are losses, so one
cannot expect to find many such lines in such a book. Further, I would imagine
many jigos could be quite peaceful short games, and such lines are not possible
to have in a book with komi 7.5 because they would be crushed by white.
My guess is simply that the tournament was played with books that leads to
fights where jigo is unlikely to happen. And in order to make 7.0 komi really
matter a lot of work for the books needs to be done.
Quoting Darren Cook <darren at dcook.org>:
>> Congratulations to Zen, winner of yesterday's KGS bot tournament!
>> My report is at http://www.weddslist.com/kgs/past/70/index.html .
> Just looking at the three strongest programs:
> zen9: 4 losses as black, 2 losses as white (24 wins)
> pachi2: 6 losses as black, 2 losses as white,
> 1 draw as black, 1 draw as white (20 wins)
> EricaBot: 6 losses as black, 3 losses as white,
> 1 draw as white (20 wins)
> Seems to (weakly) suggest that komi of 7 still favours white (*). But I
> think we need a few more tournaments for people to fine-tune their books
> and jigo handling.
> *: Which is quite possible. Little Golem still uses 5.5pt komi, and the
> results among top players shows only a very slight lead for black.
> Darren Cook, Software Researcher/Developer
> http://dcook.org/work/ (About me and my work)
> http://dcook.org/blogs.html (My blogs and articles)
> Computer-go mailing list
> Computer-go at dvandva.org
Computer-go mailing list
Computer-go at dvandva.org
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Computer-go