Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Crafty Book Implementation

Author: Robert Hyatt

Date: 13:07:32 06/12/01

Go up one level in this thread


On June 12, 2001 at 09:16:58, José Carlos wrote:

>On June 12, 2001 at 08:50:15, Robert Hyatt wrote:
>
>>On June 12, 2001 at 06:00:44, David Rasmussen wrote:
>>
>>>On June 11, 2001 at 20:48:19, Robert Hyatt wrote:
>>>
>>>>On June 11, 2001 at 14:17:37, David Rasmussen wrote:
>>>>
>>>>>Am I an idiot, or does that Crafty book scheme miss some transpositions, because
>>>>>of it's clustering system? Wait, those two statements aren't mutually
>>>>>exclusive...
>>>>>
>>>>>When some of the upper bits of the book hashkey, is from the parent position,
>>>>>doesn't that mean that if I'm in an unknown position, that can make a move that
>>>>>transposes into a book position, it won't, because the upper bits will only
>>>>>match the actual parent position(s) from when the book was created?
>>>>>
>>>>>I'm having an idea of another book implementation that I think is quite smart,
>>>>>but I won't reveal anything yet, primarily because I don't know if it works in
>>>>>practice, secondarily because it might not be a new idea, tertiarily because it
>>>>>might not be a smart idea at all :)
>>>>
>>>>
>>>>I _intentionally_ want to miss those transpositions.  One classic:
>>>>
>>>>1. e4 e5 Nf3 a6 Bb5.  Do you _really_ want to transpose back into the book
>>>>and play Nc6?  Or do you want to rip the bishop and be a piece up in a won
>>>>game?
>>>>
>>>>:)
>>>>
>>>>My thinking when I did this scheme was "Crafty will probably play Nc6 if it
>>>>is reasonable, and _then_ we will transpose back into book after white plays
>>>>Ba4, because that is a normal position.  But the position given above is
>>>>"not normal" and I won't try to transpose.
>>>>
>>>>It has happened in _real_ games in the past.  But it won't happen to me
>>>>here. :)
>>>
>>>But still, you're saying: "I'm afraid of that situationm so I won't do _any_
>>>transpositions from non-book positions to book positions", which is not as nice
>>>as being able to do it from everywhere, while still taking care of the problem
>>>you mention. I don't have a solution for this, but don't you think a solution
>>>with those properties would be nicer?
>>
>>
>>I don't see how it is possible.  I can currently transpose from any known book
>>line to any other known book line, with zero difficulty.  The only requirement
>>is that the position before I make a move must be a known book position, and the
>>position after I make a move must be one also.
>>
>>Allowing others (ie full transpositions) is dangerous due to the given position
>>where Nc6 is book _and_ BAD.
>
>  If you allow transpostion from non-book positions, you can get the behavour
>you discribed. But if you don't, you have to face the possibility of a human
>player transposing to drive you into a trick.
>  I'm at work, so I can't figure a real example, but imagine that, in the
>example you gave, axb5 was wrong due to a deep trick. A smart GM could drive you
>into that position, and the Crafty would (possibly) take the bishop happily.
>  So, the way you do things, you'll take the piece and win easily most of the
>times, but can fall into a trick, and lose badly in an important game.
>  Allowing trasposing always, you'll miss those easy wins (but you'll probably
>win anyway, since a player that gives a bishop for free will always lose against
>Crafty) but avoid the GM trick.
>  Of course, your cluster stuff is very good from a programming point of view.
>
>  José C.


I don't worry so much about the "tricks".  What I do worry about is a good
GM making a clever transposition into an opening that I don't like.  IE if he
can bypass my "don't play stonewall" opening moves, and keep crafty in book,
then it will happily play down the stonewall lines.  But if it drops out of
book for a minute, it will _never_ voluntarily walk into a possible stonewall
opening.

That is one of several reasons why I do it like I do...



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.