Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Testposition of Matador-Alarm CCT6

Author: Anthony Cozzie

Date: 06:45:40 02/05/04

Go up one level in this thread


On February 05, 2004 at 07:21:12, Tord Romstad wrote:

>On February 04, 2004 at 14:40:21, Tim Foden wrote:
>
>>On February 04, 2004 at 14:15:19, Tord Romstad wrote:
>>
>>>On February 04, 2004 at 13:07:55, Tim Foden wrote:
>>>
>>>>Some other engines I have here that will give a score for the current position
>>>>give:
>>>>
>>>>Crafty 17.11:   0.03
>>>>Yace 0.99.50:   0.09
>>>>Gothmog 0.4.5: -0.75 (well it actually says -75, but I guess centipawns?)
>>>
>>>I never thought anybody would discover the undocumented "evaluate" command
>>>in Gothmog.  :-)
>>
>>Thus we see the power of "strings gothmog.exe > tt.txt; notepad tt.txt".
>>Its a bit easier in GLC, you just start the program at a command prompt and type
>>help. :)
>
>I should probably add a "help" command to Gothmog as well.  Just the
>thought of experienced programmers like you running "strings" on my
>engine and seeing all the silly debug output strings and similar stupid
>stuff makes me feel embarrassed.  :-)
>
>>>Actually the unit is "binary centipawns", meaning pawn=128.
>>
>>Ah yes (picture of hand slapping head) I remember you saying this before.
>
>No reason to slap your head -- I never remember this kind of details about
>other engines.
>
>By the way, the scores Gothmog reports to the GUI are always translated
>to ordinary centipawns.
>
>>> In other
>>>words, the position under discussion is evaluated as -0.59.  My most
>>>recent development version is slightly more optimistic, and evaluates
>>>the position as -0.52.
>>>
>>>>So maybe I have to look a bit more into how GL gets its huge 1.422 for the
>>>>current position.  Maybe others could have a look at this and see if they think
>>>>it's complete garbage?  :)
>>>>
>>>>You can get a complete dump of the evaluation score for a position in GL by
>>>>setting a position (with the setboard or fen commands) and typing "sc".  Here is
>>>>GL's take on the root position in question:
>>>
>>>This is a cool feature.  I should probably think about implementing
>>>something similar.
>>
>>I do it with some macros, and a file that includes itself, so I have 2 versions
>>of the eval compiled into the program, both from the exact same source (except
>>for the changes in the macros of course).
>
>That's how I would do it, too.
>
>Tord

and that is how Zappa does it also.  I have found hundreds of eval bugs due to
this.  I also can check whether my debug sum matches my direct sum, this finds a
lot of bugs as well.

anthony



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.