Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: SSDF and question for Tony

Author: Tony Hedlund

Date: 06:08:55 07/24/01

Go up one level in this thread


On July 24, 2001 at 08:22:08, Harald Faber wrote:

>On July 24, 2001 at 07:48:06, Tony Hedlund wrote:
>
>>On July 24, 2001 at 05:17:48, Harald Faber wrote:
>>
>>>
>>>For Tony: Do you reset the weights in the opening book before you start new
>>>matches or do you take over the learned move preferences from former matches?
>>
>>I don't reset the opening book.
>
>
>Aha! This is one of the most important statements I have read since months.
>
>
>>But for the matches I now play it doesn't matter. In the first match, Deep Fritz
>>- Tiger 14 CB  20-20, DP was on my "left computer". In the present match,
>>Gambit Tiger 2 CP - Deep Fritz 18-21(so far), DP is on my "right computer". So
>>Deep Fritz didn't learn anything from Tiger 14 CB. But in the next match, Gambit
>>Tiger 2 CP - Tiger 14 CB, both programs have had the possibility to learn from
>>Deep Fritz.
>
>
>It is not only that DeepFritz cannot have learned from Tiger 14 to use it for
>the current match, but DeepFritz *already* has learned from former opponents.

Not in this case, since I play with new computers.

>So if you play DeepFritz (or any other program) which has learned from say 100
>games, versus a new entry, the new program definitely has a disadvantage because
>it has no learned values.

This statement *I* find interesting. Usually we get criticism because we play
against programs with *no* learner.

>>So in that sence we honours the best learner, which probably also is
>>the best program.
>
>
>Uhh, IMO this is nonsense. Oh, sorry, yes, it is the best *program*, but not the
>strongest *engine*.

I'm glad that we agree on SSDF testing for the best program.

>I don't deny that learning also counts to a program like the
>opening book, but at the moment I see a much too high value for effective
>bookleraning. One might fear that in near future several programmers will
>concentrate on *very* effective booklearning (Hi Christophe, how far is yours?)
>instead of improving the engine. Please prove me wrong...

If we just had tested engines over the years, without books and learners (and
not counting doubles). I don't think we would have such strong programs as we
have now. IMHO.

Tony

>>Tony



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.