Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Latest Carfty Se happenings

Author: Robert Hyatt

Date: 14:21:36 10/06/04

Go up one level in this thread


On October 06, 2004 at 15:46:10, Matthew Hull wrote:

>On October 06, 2004 at 15:25:16, Peter Skinner wrote:
>
>>On October 06, 2004 at 14:52:46, Peter Berger wrote:
>>
>>>On October 06, 2004 at 13:46:54, Robert Hyatt wrote:
>>>
>>>>On October 06, 2004 at 10:45:46, Tony Hedlund wrote:
>>>>
>>>>>On October 06, 2004 at 08:32:20, Mike Byrne wrote:
>>>>>
>>>>>>Alpha Pocket Crafty 19.17 SE in alpha stage.  It is playing within a GUI , will
>>>>>>have most ofthe SE features.
>>>>>>
>>>>>>Personality Opening Books -  Plan is now to generate opening books that will
>>>>>>just about every move the "real" personality played in their games for about the
>>>>>>first 20 - 25 moves.   Of course, you have to play what their opponents played
>>>>>>for it to stay in the personality book.  Fischer personality book will be
>>>>>>completed first.  It is very cool. It really give you a sense of the openings
>>>>>>the personality played.
>>>>>
>>>>>Hi Mike,
>>>>>
>>>>>In the Arena Event Forum they say that 19.17 is weaker then earlier 19.xx
>>>>>versions. Is that true?
>>>>>
>>>>>Tony
>>>>
>>>>I do not believe that, no.  The changes from earlier 19.x versions are so
>>>>small...
>>>>
>>>>you will still find some that say fritz 5 is the best, that crafty 15.x is
>>>>better than new versions, etc...  lots of opinions.  I believe the current
>>>>crafty is the best one so far, personally...
>>>
>>>At least when it is about posted results of 19.15 and 19.17, there is some
>>>evidence that 19.17 is weaker.
>>>
>>>Although I didn't collect them I realized that 19.17 was consistently doing
>>>worse in all results posted.
>>
>>
>>I think the main reason for that is flawed testing, testing on 1 pc with ponder
>>on....
>>
>>Most people do not use the correct books, proper rc file settings, and well just
>>about anything else that can go wrong.
>>
>>We have seen it many times on this forum the horrible testing systems people
>>use, and just how flawed they are.
>>
>>Crafty 19.15 - 19.17 is almost the same engine. I don't see how it is possible
>>that 19.15 would be stronger.
>>
>>From main.c:
>>
>>*   19.15   fix to outside passed pawn code that requires pawns on both sides *
>> *           of the board for the side with an "outside passer" or "outside    *
>> *           candidate" to avoid some bizarre evaluations. sel 0/0 now works   *
>> *           without crashing Crafty.  this would fail in previous versions as *
>> *           the hash signature would be modified but not restored.  slightly  *
>> *           more conservative limit on using null-move search to head off a   *
>> *           few notable zugzwang problems was added.  fix to time control     *
>> *           code to remove a hole that could cause a divide-by-zero at a time *
>> *           control boundary.  Stonewall detection removed completely as it   *
>> *           appears to be no longer needed.  rook scoring changed to better   *
>> *           evaluate "open files" by measuring mobility on them.  complete    *
>> *           removal of Phase() (phase.c) and references to the opening,       *
>> *           middlegame and endgame phases as they were no longer referenced   *
>> *           anywhere in the code.                                             *
>> *                                                                             *
>> *   19.16   fix to "Trojan code" to eliminate the time limit exclusion since  *
>> *           many users still have old and slow hardware, and the time limit   *
>> *           was not set correctly when PreEvaluate() was called anyway.  the  *
>> *           code to display fail-high/fail-low information was cleaned up so  *
>> *           that the +1 or +3 now makes sense from the black side where the   *
>> *           score is really going down (good for black) rather than showing   *
>> *           a +3 fail high (when Crafty is black) and the score is really     *
>> *           going to drop (get better for black).  Now the fail-high-fail-low *
>> *           +/- sign is also relative to +=good for white like the scores     *
>> *           have been for years.
>
>
>>adjustments to pawn evaluation terms to improve the scoring balance.
>^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
>
>It seemed that after this change, Crafty played some very good games on ICC,
>trading some good punches with Blargh and other Shredder accounts.  And, the
>ProDeo accounts could hardly score a win at all.
>
>
>
>>"new" now terminates parallel       *
>> *           threads (they will be re-started when needed) so that we don't    *
>> *           burn CPU time when not actually playing a game.                   *
>> *                                                                             *
>
>
>>19.17   changes to pawn evaluation to limit positional scores that could get a bit out of sane boundaries in some positions.
>^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
>
>I'm thinking that this change coincided with a clearing of learning files on ICC
>because crafty took a ratings dive for both standard and blitz games.  It has
>slowly regained only some of that ground.
>

Part of the "ratings dive" is external to Crafty.  IE I have been making short
runs of a computationally intense program, and occasionally someone starts a
game after I have checked to see if Crafty is playing.  And as a result, it
plays a game without getting both cpus 100%.  With my SMP code, this is always
going to kill performance....

I don't think the pawn scores really affect "normal games" much at all, they
affected games with a severe imbalance in pawn structure.  IE one side had 6
isolated pawns so that giving a 7th would produce way too big an additional
positional score, etc...  Real games don't see that very often...;



>
>
>
>> *                                                                             *
>> *******************************************************************************
>>
>>Really there is the removal of useless code, and fixs to scoring. That is about
>>it. Nothing significant between the two.
>>
>>Peter



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.