Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Testing positional "learning"

Author: Dann Corbit

Date: 23:50:56 02/04/02

Go up one level in this thread


On February 05, 2002 at 02:39:09, Jouni Uski wrote:

>Last weekend I tested positional learning under Fritz7. I played 3 engines from
>10 Nunn opening positions against Fritz7. I repeated this 4 times and checked
>their score. Result: no gain in score in any engine! Yes I checked, that lrn
>files get bigger and bigger after each match. Book learning was disabled =
>definitely Fritz wasn't learning. Seems, that this learning stuff is only waste
>of time. But I played with 60/15 time limit and ponder off - may be
>this explains my result?!
>
>Jouni
>
>PS. tested engines were Shredder5.32, Goliath Light1.5 and Yace 0.99.56

Imagine if both engines learn...
Engine 'A' says, "Here is where I went off last time..."
Engine 'B' says, "There is where I went off last time..."

Why would you expect either 'A' or 'B' to do better against the other.

There are an awful lot of possible routes from a given starting position.

Let them play 1,000 games with learning on and I suspect you will see a
significant difference.  Against identical programs that cannot see the learn
files, that is.





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.