Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: More compact EGT information

Author: Anthony Cozzie

Date: 06:29:18 04/17/04

Go up one level in this thread


On April 17, 2004 at 04:44:12, GuyHaworth wrote:

>
>Certainly, it should be possible to 'compact' an endgame's EGT information if
>higher-level rules could be found.
>
>These could preface the classic data-level EGT, providing sufficient information
> (whatever that means) without recourse to the EGT itself.
>
>In some situations, maybe the following is possible:
>
>a)  if your position is one of these, here is the value/depth
>b1) otherwise, if conditions C11, ... C1,n1 apply, use Rule 1
>b2) ...
>
>
>An example in this vein is for the DTZ50 EGTs.  Marc Bourzutschky recently
>observed that, if there are no 50-move affected subgames, we don't need a DTZ50
>EGT separate from the DTZ EGT.  We just improve the DTZ50 probe-code
>intelligence to understand that DTZ > 50 ==> draw.
>
>This saves quite a bit, even though we plan to use an 'EdZ50Z' EGT in
>association with a DTZ EGT, rather than a DTZ EGT and a DTZ50 EGT [because it's
>only necessary to encode the difference between DTZ50 and DTZ.]
>
>
>I have more than once thought that straight bit-level values ought to be useful
>somewhere.  For example, after you know you have a win, you only need 0/1 for
>'draw/win'.
>
>
>Last thought is on compression.  Kadatch's scheme is good but maybe there are
>more effective, more data-sensitive compression schemes.
>
>I gather Eugene's index-regime uses a lot of RAM.  A more 'runtime' approach to
>the maintenance of relevant indexes would I think save lots of RAM here.
>
>
>g

Actually, it seems like the most obvious thing Eugene could do is simply lose a
little precision.  Right now he is using 2 bytes / entry for most of the 6-man
tables.  He could simply throw away a little precision at the high end, and
halve the size of his tables easily.

For example:

Right now n:0-127 -> n moves to mate.

Obvious remapping: n:0->64 -> n moves to mate, n:65->90 -> 4n-(3*65) moves to
mate, n: 90->115 -> 8n-(C) moves to mate, etc.

So now his upper cap could easily be be mate in 200 or so, and the total size of
the tables is halved.  Of course, he still needed to generate the full tables,
but I might consider doing this once he finishes.  The full set of 6-men might
be under a TB then.

anthony



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.