Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: questions about book learning

Author: Robert Hyatt

Date: 19:19:22 01/04/04

Go up one level in this thread


On January 04, 2004 at 13:39:37, Bob Durrett wrote:

>On January 03, 2004 at 23:46:10, Robert Hyatt wrote:
>
>>On January 03, 2004 at 21:22:51, K. Burcham wrote:
>>
>>>
>>>
>>>I am not sure how this book learning works for each book move.
>>>The value that is changed, is this only because of the loss?
>>>Is the program using positional values once out of book to change the learned
>>>value?
>>
>>Here is a thumbnail of what I do.
>>
>>1.  Crafty remembers the evaluations for the first 10 moves out of book,
>>after each search has been completed.  It uses these evaluations to detect
>>a "trend".  IE is the evaluation good and getting better?  Is it bad and
>>getting worse?  Is it good but dropping (ie it grabbed a gambit pawn and
>>is beginning to see that it was bad) or is it bad but getting better (IE it
>>offered a gambit, the opponent took it, and the score is going up).
>>
>>It factors all of that together and marks the book line as good or bad.
>>
>>2.  Crafty takes the result of a game when it loses, and updates the book
>>line so that moves tried near the end of the line simply don't get played,
>>and alternatives near the front of the book line get tried next.
>>
>>There is more to it than that, and you can look at the crafty.doc file to
>>at least see what I am doing in more detail.  It is _very_ effective.
>>
>>
>>
>>>Lets say e4 is the most solid, dependable first move.
>>>Lets say we lose 25 games with e4.
>>>now we have a learned value for something.
>>>Is this learned value only for e4?
>>>Is this learned value only for the first move, regardlesss of the line played?
>>
>>For me, the entire line gets some "learning adjustment".  The closer to the
>>first move, the "smaller" the adjustment.
>>
>>
>>>
>>>If program A plays d4 against program B and wins 25 games.
>>>If program A plays d4 against program C and loses 25 games.
>>>If program C plays d4 against program B and loses 25 games.
>>>I am not sure how this helps the high level book.
>>
>>The danger is that you are not learning about the "book", but about
>>the "engines".  That is a problem.  Fortunately, it is not that common.
>>
>>
>>
>>
>>>
>>>Is all this learning for only the first book move?
>>
>>No.  Crafty learns for _all_ moves along a line that was played.
>>
>>In fact, if you beat it with a line, you can expect to play against that
>>line when you change colors and play it again.  :)
>>
>>
>>
>>>
>>>kburcham
>
>Bob H:
>
>I get the impression that the communication with the book, and it's modification
>is done by a GUI [or UI].  Does this imply that you have modified somebody's GUI
>[or UI]?
>
>Bob D.

Nope.  I have specific code in Crafty.  IE LearnBook() which is called by
the Crafty UI after each non-book move is played...  And at the end of the
game when the "result" is known.




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.