Author: Robert Hyatt
Date: 07:06:05 10/06/99
Go up one level in this thread
On October 06, 1999 at 08:50:57, Bernhard Bauer wrote:
>On October 05, 1999 at 10:55:37, Robert Hyatt wrote:
>
>>On October 05, 1999 at 05:45:55, Bernhard Bauer wrote:
>>
>>>On October 04, 1999 at 10:34:55, Robert Hyatt wrote:
>>>
>>>>On October 04, 1999 at 10:14:16, Bernhard Bauer wrote:
>>>>
>>>>>
>>>>>>>
>>>>>>>Game 17 looks odd:
>>>>>>>1. d4
>>>>>>>{White mates} 1-0
>>>>>>>
>>>>>>>
>>>>>>>So my impression is that Crafty is stronger than LG2000a.
>>>>>>>Kind regards
>>>>>>>Bernhard
>>>>>>
>>>>>>
>>>>>>There are some subtle places where timing can screw up. IE take this:
>>>>>>
>>>>>>white wins (crafty) and the first thing it does is to go adjust the learning
>>>>>>stuff (maybe) after playing the move that leads to mate. Then black realizes
>>>>>>that he is mated and tells xboard, which promptly declares the game a win for
>>>>>>white and tells both to start a 'new' game. Meanwhile, white has finished the
>>>>>>learning cycle, and now tells xboard that it won. IE in _any_ winboard/xboard
>>>>>>tournament, you need to examine the pgn file for each game to be sure that the
>>>>>>result given was correct. This is a problem that is not easy to fix. It might
>>>>>>be better to use the -xreuse option so that both programs get restarted after
>>>>>>each game, which means the communication pipes get reset (hopefully) before each
>>>>>>game to prevent this timing glitch.
>>>>>>
>>>>>>Else it would be better if xboard declared the wins/losses/draws and totally
>>>>>>ignore a program's claim, to avoid this...
>>>>>
>>>>>Good to know, so I will use -xreuse for future matches. BTW, on a decent system
>>>>>(linux, winnt, ...) ponder=on should be the better way to run such a tournament.
>>>>>
>>>>>Kind regards
>>>>>Bernhard
>>>>
>>>>
>>>>Note that I haven't studied this carefully... the -xreuse option might still
>>>>suffer from timing oddities for all I know. But it seems that it should be a
>>>>bit safer. Try playing a 100 game match, game in 1 minute, then check the
>>>>pgn file xboard produces to be sure there are no 1 move games... if it works,
>>>>or fails, report back so we will know whether the raw results can be trusted
>>>>or if the pgn needs inspection for all such matches...
>>>>
>>>>this will raise interesting questions about all winboard/xboard matches, and I
>>>>wonder how the chessbase winboard interface handles this?
>>>
>>>Ok, I ran that tournament. Here are the results.
>>>
>>>Third match against LG2ka
>>>Time: Game in 1 min
>>> winboard -mg 100 -xreuse -fcp "wcrafty hash=24M hashp=4M cache=2M egtb"
>>> -scp "goliath\LG2000a.exe"
>>> -tc 1 -inc 0 -sgf matchfile.pgn
>>>
>>>Legend:
>>> 0 Crafty lost the game
>>> 1 Crafty won the game
>>> d draw
>>> N game doesn't exist, not in game file
>>> X Crafty mates in 1 move
>>>
>>>Result:
>>> games 1-50
>>> 12345678901234567890123456789012345678901234567890
>>>Crafty d0X0X0X1d0X0X1110Nd10N10X0X0Xd10Xd0N11d0X0X0X0X0X0
>>>
>>>So Crafty had 18 losses
>>> 9 wins
>>> 6 draws
>>> 14 mates in 1
>>> 3 games doesn't exist
>>>
>>> games 51-100
>>> 12345678901234567890123456789012345678901234567890
>>>Crafty Xd10X1110N0N0N1110X0Xdd10N11d1d10N0N0N10X0X0X0X1d1
>>>
>>>
>>>So Crafty had 14 losses
>>> 15 wins
>>> 6 draws
>>> 8 mates in 1
>>> 7 games doesn't exist
>>>
>>>together:
>>> Crafty had 32 losses
>>> 24 wins
>>> 12 draws
>>> 22 mates in 1
>>> 10 games don't exist
>>>
>>>Looks like the results have to be treated with care.
>>>Only crafty wins in one move and only when crafty has the white pieces (22
>>>times).
>>>The game is only missing when lg2ka has the white pieces (10 times).
>>>Oh, I forgot, the match was run under WinNT 4.0
>>>
>>>Kind regards
>>>Bernhard
>>
>>
>>check the -xreuse option... that was from memory. I would hope that if the
>>programs get re-started that this would not be possible. One more thing you
>>could do if you have time: run this with -debug on winboard/xboard and send
>>me (via email) the resulting output. I might find a way to prevent this from
>>happening on my end if possible, although I am not sure it is fixable if the
>>-xreuse worked as expected...
>
>I did another test using reuse and reuse2:
>
>winboard -mg 50 -reuse -reuse2
>-fcp "wcrafty hash=24M hashp=4M cache=2M egtb"
>-scp "goliath\LG2000a.exe" -tc 1 -inc 0 -sgf matchfile.pgn
>
>but the result was the same as before, missing rounds and mates in one.
>So I will run a test with -debug, but that will happen over night, so nothing
>new now.
>Kind regards
>Bernhard
Then this might simply be a 'race' condition between crafty and xboard. It
sends a "new" command with reuse enabled, or it sends a 'quit' with reuse
disabled. Crafty sends a result message at the end of every game. It is
possible that the result passes the quit in the pipe and xboard sees it
anyway. Perhaps..
I could add an option to turn off claiming a win, but then I don't think that
games will terminate cleanly...
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.