[Top][All Lists]

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

Re: [Bug-gnubg] gnubg "personalities"

From: Douglas Zare
Subject: Re: [Bug-gnubg] gnubg "personalities"
Date: Mon, 10 Jul 2006 08:24:41 -0400

On 7/9/06, Joseph Heled <address@hidden> wrote:
 I may be
wrong but offhand I can't remember any of Douglas ideas that were
eventually implemented in gnubg.

That may be because, as I complained before, people have done a poor
job of attributing ideas. See


which, btw, was prompted by multiple requests for help analyzing
rollouts truncated at a z-score threshold (# standard deviations).

I have written no code for gnubg, but here are three topics that have
been implemented in gnu in part due to my work:

1) The effective pip count. Of course Walter Trice had the idea before
me, and wrote some articles in GammonVillage in his Beginners'
Bootcamp series, around 17-19 IIRC. If you would like to say you got
the idea from Walter instead, fine, make sure to acknowledge Walter.
However, I think it was after I wrote a series of articles for
GammonVillage and posted analyses of positions on Gammonline and
rec.games.backgammon, some included in the rec.games.backgammon
archive, that the idea was brought up here by people citing my work.

2) "Casual player" and the threshold of 3.5 mppm for world-class play.
Suggesting the term is not very important, but it always jumps out at
me when I see it. I also argued for using a stricter cutoff for
"world-class" than Snowie's 4.4, and I recall tabulating the results
of several actual live tournament matches from GammonVillage to show
that an error rate of 3.5 is attainable by some good players.

3) Luck-adjusted match/session statistics. The idea of unbiased
variance reduction is not originally mine, even within backgammon, I
learned about it by reading David Montgomery's description of what was
implemented in Jellyfish before I started playing backgammon
seriously. However, I believe the idea of applying unbiased variance
reduction to a match between two players was introduced by me in one
of my pre-column GammonVillage articles (as one of the things I would
most like to see in a bot). I mentioned it several times elsewhere,
such as when I discussed wanting to apply unbiased variance reduction
to the JF vs. Nack and Senk challenge, and when I discussed formats
for a meaningful computer backgammon tournament. I believe several of
those references were cited here before the luck-adjusted statistics
were implemented.

I'll be travelling for the next 10 days, so I will be out of regular
e-mail contact.

Douglas Zare

reply via email to

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