Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Vincent

Author: Tord Romstad

Date: 05:02:23 05/07/04

Go up one level in this thread


On May 07, 2004 at 07:09:54, Sune Fischer wrote:

>On May 07, 2004 at 06:32:12, Tord Romstad wrote:
>
>>On May 07, 2004 at 04:10:01, Sune Fischer wrote:
>>
>>>Ie. if Movei could score 40% against Crafty I wouldn't be surprised. Gothmog has
>>>scored 50% in some tests I believe,
>>
>>You are probably referring to some of the match results I have reported
>>myself.  The last couple of times I have played matches against Crafty,
>>the scores have been almost exactly 50%.  But it should be pointed out
>>that the match conditions are not very fair to Crafty, for two reasons:
>>
>>1. I play my matches on a single-CPU machine, with pondering disabled.  As
>>   is well known, Crafty is seriously crippled without pondering, because
>>   its time management doesn't work well.
>
>Perhaps it is just Gothmog that is crippled with ponder on? :)
>I imagine that you, like me, hasn't spent a lot of time testing with ponder on
>because we don't have dual machines to test on.
>
>In any case if an engine is performing like the author designed it then it's
>fair game.
>
>Anyway, what is different in Crafty's TM with ponder on?

I don't know the details; we should probably ask Bob.  I just recall
numerous occasions in the past when somebody reported a bad match result
for Crafty and Bob explained that the match result didn't mean anything,
because Crafty was not designed to work without pondering.

For this reason I don't test against Crafty very often.

>>2. My Crafty executable is compiled using GCC on a Linux machine.  The
>>   Windows version of Crafty is probably compiled using the Intel or
>>   Microsoft compiler (unlike Gothmog, which uses GCC on all platforms),
>>   which reportedly produce much faster code.
>
>Ok, but using a better compiler is not really a strength of the engine. I think
>using the same compiler makes a good basis for comparison since Gothmog would
>probably get the same kind of boost with a better compiler.

No.  Gothmog probably wouldn't even compile with another compiler.  I don't
really know the C programming language.  I program in a mix of ANSI C,
Unix-specific functions, and some GNU-specific extensions, and I have no
idea which of the many functions I use are portable.  In theory it would
of course be possible to port my engine to work with the Intel or Microsoft
compiler, but it would probably require some non-trivial work.

Bob has already done this work.  When comparing Gothmog and Crafty as
Windows engines (which is what the majority of people care about), it is
perfectly fair that Crafty benefits from a better compiler.

>>I also think it is possible that Gothmog performs better against Crafty
>>than against other engines of similar strength, for reasons of playing
>>style.  Crafty's lack of king safety eval makes it rather vulnerable to
>>Gothmog's very aggressive play.
>
>I can't figure out how strong Gothmog is.

Neither can I.  It appears to be in the same league as Frenzee and Movei,
but on the other hand it also scores around 50% against Yace Paderborn.  It
seems to be hundreds of rating points weaker than Amateur, The Baron and
Butcher.

Of course, this doesn't add up at all.

>Sometimes it appears almost to be at the Ruffian level, other times it has
>problems with more mediocre engines.

Yes, exactly.  Confusing, isn't it?

>Perhaps it really does have something to do with style, or simply too few games
>played and it really is somewhere inbetween.

Gothmog is a very unbalanced engine.  It is quite good in some respects, and
horribly bad in others.  The search and eval is fundamentally (and
intentionally!) unsound, in an attempt to make the engine play an interesting
game.  This works very well in some cases, but sometimes has catastrophic
consequences.

I think it is relatively easy to tune any reasonably strong engine to score
almost 100% against Gothmog.  Ask Anthony if you want to know how.  :-)

Tord




This page took 0.01 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.