Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Why do programs have such a hard time with pawn promotion threats?

Author: Pete R.

Date: 15:51:40 08/05/99

Go up one level in this thread


On August 05, 1999 at 13:14:13, José de Jesús García Ruvalcaba wrote:

>On August 04, 1999 at 19:03:25, Pete R. wrote:
>
>>On August 04, 1999 at 14:57:02, José de Jesús García Ruvalcaba wrote:
>>
>>>On August 03, 1999 at 21:29:14, Pete R. wrote:
>>>
>>>>Currently in the Kasparov vs. The World correspondence game there are a number
>>>>of endgames that involve races between passed pawns on either side of the board.
>>>>While the programs have been quite useful so far, they aren't too swift in the
>>>>transition to the endgame when passed pawns are the issue.  Where a human player
>>>>would push his passed pawn at every opportunity, the programs tend to diddle
>>>>around moving pieces, until the pawn is almost promoted, when they suddenly
>>>>realize that one side is winning. For example in the following position,
>>>>
>>>>8/1p1kp2Q/2np4/1p2bpB1/8/3q3P/5PP1/5RK1 w - -
>>>>
>>>>I ran Fritz 5.32, Hiarcs 7.32, and Crafty 16.12 for many hours, enough to get to
>>>>18 ply for Fritz and Crafty.  All of them recommend the move g4 for white
>>>>(Crafty switches to g4 at 15+ ply), which gets buried by b4!.  For example,
>>>>
>>>>27. g4 b4
>>>>28. Qxf5 Qxf5
>>>>29. gxf5 b3
>>>>30. Rb1 b2
>>>>31. Kf1 Na5
>>>>
>>>>and by now they will realize that white is losing.  There are a number of
>>>>possible endgames coming up involving a race between white's passed h pawn and
>>>>black's passed b pawn, and the programs have a real hard time with it.  What's
>>>>going on here?
>>>
>>>	I recommend to use crafty 16.15 with 'extension pushpp 1.0' for this and
>>>similar positions, it will extend more aggresively on passed pawn advances (the
>>>default is 0.75), and will see the dangers at shallowers depths.
>>>	I can not do it myself because I do not have "many" hours of computer time
>>>available.
>>>José.
>>
>>Thanks, I'll play with these new parameters.  I think the ppscale might be the
>>one to use vs. the extension, but I'll have to fiddle with it.
>
>	Yes, increasing 'evaluation ppscale' can also help. Values over 200 are not
>recommended (the default is 100). I love these features of crafty.

Actually I couldn't get the extension command to work.  I put it in the rc file,
and no matter what syntax I tried I got an error - unrecognized parameter
message.



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.