Usenet.com

www.Usenet.com

Group Index

Rec Thread Archive from Usenet.com

<-- __Chronological__ --> <-- __Thread__ -->

Re: BGBlitz 1.7 won against GNU-BG 0.14 at the 8 Computer Olympiad.....



Douglas Zare <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...
> Frank Berger wrote:
> 
> > BGBlitz 1.7 won against GNU-BG 0.14 at the 8 Computer Olympiad!
> > It was a best of five 13pt match.
> 
> Congratulations on your victory.
Tnx :)
 
> What do you think the odds are that BGBlitz wins in a 13-point
> match against gnu or Snowie, and would you be willing to back
> that assessment for money?

I havn't analyzed the matches deeply so far and I hope Achim will
provide a thorough analysis. My first impressions is that the checker
play is extremely well. The net used in Graz is about 1,5% better in
cubeless play than the version 1.6. Using 3-Ply adds further
improvement. Thesauro writes, that between TD-Gammon 2.1 and 3.0 80%
of the strength increase was by using 3-ply, but in the case of
BGBlitz I think this number would be too high. I guess adding another
1,5% is a lower bound.

The cubeplay especially in matches has to be improved and this will be
my next main activity (btw one resource will be your articles ;) ) for
the 1.8 release.

So what are the odds? I guess in cubeless play, there is less than 1%
difference to GNU (I don't have access to Snowie to judge that)
in whatever direction. It was amazing at Graz, when sometimes Achim
and me looked at each other saying: "this move can't be right", just
to see that both bots agree. Solidarity beyond silicon brains?

In matchplay.... hard to tell. Probably I have to ask an mathmatican
;)
Currently the cubing algorithm in matchplay neither uses volatility
analysis nor cube vigorish and the matches in Graz showed that all but
one error were cubing errors. I'm quite confident that I will come up
with an state-of-the-art cubing soon and then it will be as close as
with the checkerplay.

In terms of your offer for moneyplay. Believe it or not: I never
played a single game of Backgammon for money. I will not change that,
just to prove anything. And naturally you know as good as I, that such
a game will be totally determined by random. Otherwise I'm open to any
challenge, e.g. at FIBS.

BTW may I ask for the status of your bot? I'm very curious about it.
Don't forget my offer, BGBlitz is able to host several bots, so if you
want to take a shortcut.....

ciao
Frank



<-- __Chronological__ --> <-- __Thread__ -->


Usenet.com



Please check out one of the premium Usenet Newsgroup Service Providers below for access to Usenet.