Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Interesting, But Do These Results Tell The Whole Story?

Author: Uri Blass

Date: 00:55:09 05/12/01

Go up one level in this thread


On May 11, 2001 at 22:25:07, Steve wrote:

>Thank you for posting these very interesting results.  But I wonder if these
>kinds of tests truly measure how well a program analyzes.  I own Hiarcs 7.32,
>and if you give Hiarcs 2 minutes to analyze a position and Program X 2 minutes
>to analyze a position, Program X may do better.  But in actual game analysis,
>the fact that Hiarcs analyzes backwards (like all ChessBase products) and stores
>hash tables between moves may produce game analysis of much higher quality than
>Program X, if Program X has neither of those features.

I doubt it.
Hiarcs has a bug in retaining hash tables and I saw cases when Hiarcs cannot see
things when it analyze backwards when it can see them without analyzing backward
because it learn wrong information from hash tables.

This wrong learning is also a problem in games and I saw cases when Hiarcs
had not enough time to solve a fail low in games because of the bug of retaining
hash tables(It can solve it after enough time but practically it has not enough
time)

This 'enough time' can be hours if you play games at slow time control like 4
hours/40 moves.

It seems that the programmer of Hiarcs only tested it in blitz so he missed bugs
at long time control.

Uri



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.