Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 1. Rg3! and Fritz 9

Author: vladan

Date: 05:35:26 11/08/05

Go up one level in this thread


On November 07, 2005 at 09:53:40, Djordje Vidanovic wrote:

>On November 07, 2005 at 08:29:00, vladan wrote:
>
>>On November 06, 2005 at 21:18:21, Djordje Vidanovic wrote:
>>
>>>If you take a look at this position you can see that Fritz 9 needs a decent
>>>chunk of time to find the simple Rg3! It spends a lot of time on the simple pawn
>>>to g4 and only after about 80 seconds (FX-53) switches to Rg3, eventually
>>>announcing a mate in 23:
>>>
>>>[D] r6r/p2Q2pp/6k1/2P1pp2/1Pb5/R7/6PP/6K1 w - - 0 29
>>>
>>>
>>>It is interesting to note that Fritz 9 learns successfully, even though its
>>>positional learning is all done in the RAM and gets cleared the moment you exit
>>>Fritz.  In a repeated try while I was still in the program, Fritz targeted Rg3!
>>>right away never changing the winning line.
>>>
>>>
>>>Djordje
>>
>>Very interesing observation Djordje, but I think its not machine learning,
>>simply Fritz uses old hash generated by primary searching. Try to reset
>>hash, and then search the position again. Ruffian also has advanced hash
>>procedures to use previous search results, but it is only hashing.
>>
>>
>>Best regards,
>>
>>Vladan
>
>
>Hi Vladan,
>
>nope, Fritz 9 does learn albeit only during longer engine-engine matches that
>are played non-stop, or while at the server overnight etc.
>
>Trust me, I know what I am saying :-)
>
>
>Djordje



Hi Djordje,

you are right, Fritz 9 has adapted match learning, but there is nothing new,
many other (even free) programs has the same procedure.

The true machine learning is non-adapted (opponent indepedent learning).

Fritz 9 has not that technique, for sure.

Vladan









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.