Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: The Brick Wall

Author: Andrew Williams

Date: 12:19:15 09/20/04

Go up one level in this thread


On September 19, 2004 at 19:28:28, Stuart Cracraft wrote:

>On September 19, 2004 at 15:18:47, Andrew Williams wrote:
>
>>On September 19, 2004 at 14:21:30, Uri Blass wrote:
>>
>>>On September 19, 2004 at 12:54:57, Andrew Williams wrote:
>>>
>>>>On September 19, 2004 at 11:10:14, Stuart Cracraft wrote:
>>>>
>>>>>Hi -- I am looking for 2 or 3 beta testers who would receive
>>>>>(full) source code to my program and in return would provide
>>>>>input and comments about improving the search. They would
>>>>>simply agree not to redistribute it and in fact discard it after
>>>>>a week or two of looking at it (and commenting.) The program is in C,
>>>>>5000 lines. The search and quiescence routines are 600 lines total.
>>>>>
>>>>>The reason I am considering this is due to hitting a brick wall at 249/300
>>>>>on WAC for several weeks now and knowing there are things I just cannot
>>>>>find or go further with. The above score is at 1 second per position on a
>>>>>1ghz P3 with a small transposition table. I am told that 270-300 is considered
>>>>>"good" for this time control on this test. On this same machine
>>>>>at the same time setting, with WAC, Crafty gets 270/300.
>>>>>
>>>>
>>>>It strikes me that comparing your program with crafty (or any other program for
>>>>that matter) based on 1 second searches in WAC is a bit weird. And a waste of
>>>>time. I'm pretty sure, for example, that my program would do worse at WAC 1
>>>>second searches than yours on that hardware and I certainly don't care either
>>>>way.
>>>
>>>Did you test it.
>>>If not how can you be sure about it?
>>>
>>>Without testing my guess based on the level of your program is that you do
>>>better than 250 solutions  in WAC at 1 second per move on stuart Hardware
>>>because I expect every program that is at the level of postmodernist to do it.
>>>
>>
>>To be honest, it was a wild guess. PM gets 269 right in 1 second on my AMD 64
>>3200. Probably you're right and I was wrong to say "pretty sure". But it doesn't
>>matter at all.
>>
>>>Crafty is not relatively strong in tactics of short searches and there are a lot
>>>of weaker programs who do better than it.
>>>
>>>If a program does significantly worse than Crafty in WAC at 1 second per move
>>>then there is a lot to improve in it's search.
>>>
>>
>>That may be true, but I would reiterate that looking at its performance in WAC
>>is not going to help Stuart much in improving it. I don't even think it will
>>help much in improving its performance on other tactical tests, but that is just
>>a guess. I would strongly re-state my point: to learn what is wrong with a chess
>>program, it is better to play games than to test over and over on a test suite.
>>Even testing over and over on several test suites is not a good idea, in my
>>opinion.
>
>Ahh, but you may have forgotten: I am a *very* weak player.
>
>That would be like Fide Master David Gliksman advising Gary Ksaparov.
>
>Doesn't make sense. In fact, the difference in rating points is probably
>similar.
>

We're *all* very weak compared to our programs. That doesn't matter, because
one's ability to play is different from one's ability to identify poor play.

>I am likely around 1400 and my program is already showing over master
>the IQ test... so this just doesn't make sense.
>

I don't understand this. Are  you saying that your program is better than master
level? By what measure?

>I can understand test results. Throw me a position and I drown.
>
>Stuart

I find it hard to see what you mean by "I can understand test results". Do you
mean that you can understand when your program gets some more right than the
previous time you tested it? What are you learning, however? You reported that
crafty gets 270/300 while your program gets 249/300 on WAC/1 second. What would
be the result if you played a 100 game match between your program and crafty?

Please don't take any of this to be criticism. I like to see someone working
hard at their chess program, no matter what approach they take.

Cheers

Andrew





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.