[Top][All Lists]

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

Re: [Bug-gnubg] Holding games

From: Joseph Heled
Subject: Re: [Bug-gnubg] Holding games
Date: Mon, 23 Jun 2003 12:24:22 +1200
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.2) Gecko/20021202

I suspect that just holding an anchor would mark too many positions. Probably many you would not consider as a "holding game".

My benchmark shows that 46.5% of the positions are classified as holding by your definition. The error rate definitely supports the claim that GNUbg errs more in those positions.

nonholding (57431 53.43%) 0.009573
   holding (50054 46.56%) 0.011446


Albert Silver wrote:
I was gone this weekend so couldn't respond earlier.

I don't think we need that many to make the case, but if you can
generate some positions I'll put them through sagnubg.

I agree. Still waiting for a clear cut definition of a holding game


you guys. How about "No checkers on points lower than 18, decent


(say at least 3 points separate the back checkers), and no blots in


area of contact?

I'm not sure I understand your desription. As I see it, you have a
holding game if you have an anchor on my 3-pt, 4-pt, 5-pt, or 7-pt (my
bar point). For example:

GNU Backgammon  Position ID: 4HPDAQbg2+DBAA
                 Match ID   : cAkAAAAAAAAA
 +13-14-15-16-17-18------19-20-21-22-23-24-+     O: GNU
 | X        O  O  X |   | O                |     0 points
| X O O X | | O | | X O | | O | | X | | O | | | | O | v| |BAR| | (Cube: 1) | | | X | | | | X | | O | | X | | O X X | | X O | On roll
 | O           X  X |   | X     O          |     0 points
 +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: Albert

Here the Os have a 3-point holding game, and the Xs have a bar-point
holding game. Of course, one of the big issues (highlighted by Neil) is
that the bots tend to overestimate positions where the holder has a 3rd
rear checker such as:

GNU Backgammon  Position ID: 4LnhACPgbfBgIA
                 Match ID   : cAkAAAAAAAAA
 +13-14-15-16-17-18------19-20-21-22-23-24-+     O: GNU
 | X        O  O  X |   | O              X |     0 points
| X O O X | | O | | X O | | O | | X | | O | | | | | v| |BAR| | (Cube: 1) | | | | | | | X | | O | | X | | O X X | | X O | On roll
 | O           X  X |   | X     O        O |     0 points
 +12-11-10--9--8--7-------6--5--4--3--2--1-+     X: Albert

I wasn't too sure about what was meant with " no blots in the area of
contact". Do you mean if the blot on 1 was on the 5 point instead? If
so, it would still be a 4-point holding game, even if it could change.


Neil Kazaross has made the observation that "most bots tend to cube
holding games a bit late". If an expert of Neil's calibre has noted


feature then I'm inclined to take it seriously. One of the best ways


improve gnubg must be to identify classes of position where it is


Expert opinion will be very valuable for this.



Bug-gnubg mailing list

reply via email to

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