Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How to build EGTB?

Author: Severi Salminen

Date: 03:19:33 02/24/01

Go up one level in this thread


>>I don't quite understand your question. I don't understand what switch/... has
>>to do with EGTB? Could you be more specific - what do you mean by "normal"?
>
>I means the question: is there any other method than EGTB for getting same
>result?

Oh, now I understand. Yes, there are definitely other methods, but the result is
not exactly the same. EGTBs contain perfect knowledge. You allways have the best
moves and true scores, but they are not very convenient: lots of hard disk
accesses -> slow and they require huge amount of disk space (even 5 piece EGTBS
need 10+ GB). One very interesting method is Ernst Heinz's Interior-Node
Recognition. This has same principles but different approach:

1. Check out if we have the appropriate "recognizer" (that is a _function_ for a
certain piece configuration - not a file like in EGTBs)
2.If we do have we call that function which tries to give score to current
position.
3. If the position is recognizable we return a INR score, if not we return NULL
or something and continue normally.

An example: we arrive to a position KQK with stronger side to move. We first
conclude that we have the recognizer kqk(). We call it. The recognizer then
evaluates the position and returns a score. So in this case the score should be
higher if the losing side's king is nearer the edge. Obviously we have to
perform very quickly the initial recognizing: do we have the appropriate
recognizer. We must do this in every node so this must be optimized. But the
same applies to using EGTBs.

The benefit of using recognizers is speed and the usage of memory: all
recognizer functions can be in memory at the same time so the access times are
_a lot_ faster than when using EGTBs. The disadvantage is that they must be
programmed and they don't necessarily allways return the best move but only a
move leading to victory (this is not so bad after all). You should check out
Heinz's book Scalable Search in Computer Chess or the website containing
everything: http://supertech.lcs.mit.edu/~heinz/dt/

>>By downloading a EGTB generator from somewhere and generating the files yourself
>>(you need patience, a lot RAM and hard drive space and a fast processor) or
>>downloading the EGTB files themselves (recommended).
>
>Could you recommend me some links for downloading generator and EGTB files?

From Robert Hyatt's ftp site you can get them all (the TB directory):

ftp://ftp.cis.uab.edu/pub/hyatt/

But I really don't recommend you to build those EGTBs yourself if you can
download the pregenerated EGTBs (you need a fast connection). You can get both
the generator and the generated files from that site.

Severi



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.