Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: rebel 10~!! super strong on amd k62 500

Author: Ratko V Tomic

Date: 13:39:40 07/28/00

Go up one level in this thread


> Hiarcs is good at this time control.
> I think that the learning bug of Hiarcs becomes a problem in
> longer time control
>

Well, I have ran few times (out of hundreds of games) into this "bug" of Hiarcs,
where it played a very risky, dubious move, and in a later analysis it couldn't
reproduce it. It has clearly to do with the persistent hash. But this may not be
a "bug" but merely a tradeoff. You allow it to lose few games, but on average
you gain. I haven't seen any benchmarks between Hiarcs with different sizes of
hash tables (from 0 and up) playing against itself, but I suspect that the net
outcome would be that the table with hash would come ahead the one without hash.
Of course, you will find positions where the non-hash version computes some move
better, but this is still fine with the tradeoff hypothesis.

>I know that the difference between Junior6 and Junior5 or Fritz6 and Fritz5.32
>is a big difference in ssdf rating.
>
SSDF or similar comp-comp tests tend to exaggerate the differences (relative to
a wider pool of players). The newer programs play better against the fixed
weaknesses of the older programs. By hitting consistently into such weaknesses
they are replicating in more elaborate form of the 'killer book' theme. The
ratings will thus diverge artificially, i.e. they won't diverge nearly as much
in the play against IMs and GMs.






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.