Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: mclane's summer-tournament: end of round #8

Author: Enrique Irazoqui

Date: 05:30:09 09/30/98

Go up one level in this thread


On September 29, 1998 at 19:19:42, Moritz Berger wrote:

>The PowerBook works fine - IF LEARNING IS ENABLED.
>
>Thorsten has stated here that he uses the book on CD. So there's no learning
>taking place at all.
>
>Also, the book learning concept naturally works best when a large number of
>games against preferrably strong opponents gets played. This is usually the case
>e.g. in autoplayer (no problem at SSDF) or server games (no problem e.g. for
>Mark Young). This is also the case when you play a "match" of e.g. 20 games
>against the same opponent. That's why autoplayer and server results are better
>than "amnesiac" results without any learning going on or learning from only a
>handful of previous games, some against opponents more than 200 ELO points
>weaker than Fritz.
>
>Remedy if you don't play many games but want to benefit from learning: Let Fritz
>learn e.g. from the public SSDF games (maybe only those where also the PowerBook
>was used). There you have several hundred games, add the games from Enriques
>tournament if you like and Fritz will probably benefit more from its book.

I don’t think a learner can have any significance in this kind of tournament
where a program plays one game at a time against different opponents for a total
of 8 games. With the Powerbook on the hardisk, Thorsten would have had the same
results.

The learner in Fritz 5 works wonders in long matches played in sequence. In the
40:2 tournament I played, Fritz 5 played 4 matches of 20 games each. In the
first 10 games of each match, Fritz 5 scored 44%. In the last 10, 61%.

Looking for the same pattern in the 426 games I have from the SSDF, Fritz 5
scored 64% in the first half of the matches and 72% in the second half.

This can explain the difference in Fritz 5 between manual playing, with one game
at a time against different opponents, and autoplaying games in sequence against
the same opponent. It can also explain why I never thought Fritz 5 was stronger
than other top programs, in spite of its great comp-comp results.

Right now I am playing a match Fritz 5.04 – Rebel 10. Fritz started losing in
the fir



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.