Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Why do some engines have learning capacity and others none?

Author: Uri Blass

Date: 00:51:32 01/07/06

Go up one level in this thread


On January 06, 2006 at 02:54:44, Peter Skinner wrote:

>On January 06, 2006 at 02:42:55, Premraj Natarajan wrote:
>
>>On January 06, 2006 at 01:39:00, Peter Skinner wrote:
>>
>>>On January 06, 2006 at 00:46:40, Lonnie Cook wrote:
>>>
>>>>I always thought learning capcity for your engine was a good thing
>>>>
>>>>Those with this feature:Shredder 9,HIARCS10,Nimzo,Crafty,Comet just to name a
>>>>few
>>>>
>>>>
>>>>But then why do the majority have no learning capacity?  Like:
>>>>
>>>>Fruit 2.2.1,Rybka,Fritz,Junior,KING 3.23  to name a few.
>>>
>>>Well Fruit, Junior, Fritz, and The King 3.23 all having learning done by the
>>>interface if possible.
>>>
>>>The Chessbase interface and Chesspartner both do learning for the engine.
>>>
>>>Crafty, Comet, and Hiarcs 10 have native learning and positional learning.
>>>Shredder is still an engine that relies on the interface to do the learning. The
>>>engine itself does not learn.
>>>
>>>Some have it because the authors believe it to be beneficial, and others have
>>>not implemented it because they think that learning over time can actually hurt
>>>performance.
>>>
>>>This arguement is the same for tablebases.... it is just the way the world
>>>works.. :)
>>>
>>>Peter
>>
>>Shredder has very good learning compared any other engine mind u it learns
>>correct move for a position and writes it in a shredder.pl2 file
>>
>>Raj
>
>As i don't have Shredder 9 I just assumed the learning was the same as in 8. I
>did not know Shredder has positional learning.
>
>While that is great, to say that one program will play x move in a given
>position does not mean that another will do the same.
>
>This is where the combination of book learning, positional learning, and result
>learning are so important.
>
>Peter

It seems that all this learning is a big failure in comp-comp games(at least
against fruit).

programs in the ssdf list fail to improve their result against fruit with more
games.

I wonder if programmers are going to improve their learning in order to improve
their score in long matches against programs with no learning.

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.