Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty Book Implementation

Author: Robert Hyatt

Date: 13:15:08 06/12/01

Go up one level in this thread


On June 12, 2001 at 10:05:50, David Rasmussen wrote:

>On June 12, 2001 at 08:52:38, Robert Hyatt wrote:
>
>>
>>"Unintentionally, by search" is what I want.  I want to use the search to find a
>>reasonable move if the beginning position is not a known book position.  If it
>>happens to find a move that takes it back into book, fine.  If it doesn't, that
>>is perhaps even better (again, the Nc6 move is what I want to avoid, so that I
>>can play axb6 and win easily.)
>>
>>To answer your question "how can it hurt" just remember e4 e5 Nf3 a6 Bb5 and
>>recall that Nc6 takes you back into book.  axb5 wins the game.  Which do you
>>_really_ want to find?  :)
>
>I see all your points (in fact, I saw them before you made them :), but I
>_still_ think it must be nicer to somehow have transpositions from non-book
>positions to book positions and simultaneously avoiding the problem you mention,
>if it can be done. Because there must be some cases where Crafty chooses by
>search, to go into a non-book position, where it _doesn't_ win a piece. That is,
>the move that Crafty chose wasn't as good as any of the book alternatives that
>existed. If Crafty always in those non-book positions chose a move that turned
>out to lead to a book position, it wouldn't need a book at all, except for the
>saved time it gets.
>
>I'm not saying I know how a system such as the one I describe, can be done. But
>I will think a bit more about it, since I don't like the idea of missing book
>positions in cases where it would be a good idea not to.

It is trivial to do.  Just use the complete 64 bit hash signature, rather than
the 16+48 approach I use.  I used this in Cray Blitz, and in early versions
of Crafty.  The current approach has a couple of good points (speed, not
transposing from one line to another unless _both_ are in the book already).
The other approach is much slower (you must search for the hash signature since
there is no way to 'group' them as I do now) but will find _any_ possible
transposition.  That got us into trouble many times with Cray Blitz.  We
followed a book line to a point where we wanted to play move "X" or else go
off on our own, depending on what the opponent did.  And he would play something
that would jump us to another book line that ended up much worse than we wanted.






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.