Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: knowledge and blitz; search and long games

Author: Vasik Rajlich

Date: 07:04:28 02/15/06

Go up one level in this thread


On February 15, 2006 at 03:07:50, Joseph Ciarrochi wrote:

>I have a naive question...
>
>in my understanding, Fruit has excellent search efficiency but not huge amounts
>of knowledge. In contrast, Fritz 9 and Rybka have substantial knowledge. If you
>can trust Rybka's depth outputs, it does not seem to be as quick at getting to
>deeper plys.
>
>I have observed that Fruit 2.2.1 tends to play poorly at blitz and improve
>steadly with long time controls, with it being an absolute god on the longest
>time controls (SSDF). In contrast, both rybka and fritz 9 play blitz well.
>
>do programs with more knowledge tend to play blitz better? Knowledge is kind of
>a quick, heuristic way of making a decision about what is likely to work. It
>presumably can come into play very quickly. In contrast, search takes time.
>However, it does discover when the knowledge is not useful (i.e., when the
>knowledge heuristic is inconsistent with the concrete variations uncovered by
>search; e.g., doubled pawns may generally be bad (knowledge heuristic), but in
>some situations can be quite good)
>
>is my reasoning correct? Maybe it would help for me to understand what
>constitutes "knowledge" in a chess program. I always presume its things like
>"doubled pawns are often bad" or two bishops are good, or it is often good to
>push pawns and have space..
>
>
>best
>Joseph

We need to keep our terminology straight.

Chess knowledge (in the context of computer chess) is what makes a program play
well. At standard time controls, Fruit probably has a tiny bit more chess
knowledge than Fritz and Hiarcs.

You can also talk about the complexity of a chess program. Hiarcs is probably
the most complex of the above three, and Fruit the simplest. Shredder is another
complex program. I suspect that the more complex programs are better at faster
time controls.

BTW - one (unfortunate) way to measure program complexity is:

[program bugs or weird behaviors] * [program ELO]

Vas




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.