Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Book "learning" is strange.

Author: Anson T J

Date: 02:12:49 03/11/04

Go up one level in this thread


On March 10, 2004 at 18:27:20, Anson T J wrote:

>On March 10, 2004 at 17:52:59, Louis Fagliano wrote:
>
>>On March 10, 2004 at 16:12:51, Anson T J wrote:
>>
>>>On March 10, 2004 at 15:53:36, Louis Fagliano wrote:
>>>
>>>>Is there something wrong with Shredder’s book in game 26?  Look at the game (now
>>>>in progress at the time of this posting).  Why did it open up with 1. e4 e6 2.
>>>>Qe2 c5 3. g3 Nc6 4. Nf3 g6 5. d3 Bg7 6. h4?  A rather bizarre opening that gives
>>>>White less than nothing.  It was out of book by move 5. d3.  Is Shredder book
>>>>accidentally set on a wide variety or“fun” level or even accidentally turned
>>>>off?
>>>
>>>I assure you both books are on their strongest settings. On round 18, Junior
>>>playing black got a nice position playing the french defence. Shredder managed
>>>to draw but perhaps learnt not to play mainline French as white so went for 2.
>>>Qe2 this time around. It had a probability of 27.2.
>>>
>>>The rest of the moves are just the top lines in the Shredder8 book.
>>>
>>>http://homepage.ntlworld.com/ansonwilliams/chess/junior8test/shredder8book.jpg
>>>http://homepage.ntlworld.com/ansonwilliams/chess/junior8test/shredder8book2.jpg
>>>http://homepage.ntlworld.com/ansonwilliams/chess/junior8test/round19.jpg
>>>http://homepage.ntlworld.com/ansonwilliams/chess/junior8test/round20.jpg
>>
>>There seems to be a problem with how some of these programs "learn".  You said
>>that "On Round 18, Junior, playing Black, got a nice position playing the French
>>Defence", and now when I look at Shredder's learned adjustments it is now NEVER
>>going to play 2. d4 after 1. e4 e6.  That's ridiculous.  Whatever nice position
>>Junior got playing the French, 2. d4, could not possibly be the cause for
>>defeat!  It's probably something around moves 10-18 or so that White played
>>something inferior.  2. d4 is not most certainly not a mistake in the French!
>>
>>I noticed this among all ChessBase playing programs in the CB GUI.  Whenever
>>they win or lose and "learning" is enabled, they pick the first or second move
>>of the game to "adjust".  For example, if any ChessBase playing program loses as
>>White in the Ruy Lopez, it will cease playing 1. e4(!)
>>
>>Something is wrong here.  The ChessBase playing program should be looking much
>>further down the road in it's book, starting with the last book move and working
>>backward looking for suspected inferior book moves.  This doesn't seem to
>>happen.  Anyone else notice that?
>
>Its not as bad as you think. The deeper moves are penalized more than the ones
>at the root. Given enough "French" Shredder will return to 2. d4! If the book
>maker didn't want shredder to play 2. d3 and 2. Qe2 ?! then those moves
>shouldn't be tournament moves.

Shredder defeats Junior's French in round 28 with 2. Nc3. Now in round 29,
Junior is playing a variation of the French as white.

http://homepage.ntlworld.com/ansonwilliams/chess/junior8test/

I do agree that the Chessbase book management system is too weak. The books
themselves are okay but the user isn't given many options to utilise them well.
I think we should be able to set how much the percentage sucess counts and how
much the frequency of play counts. I believe that the frequency plays too big a
part in the selection of move.

The match is still very close and it seems Junior is fighting hard to maintain
its unbeaten record in my tests.



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.