Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question to people who generate 6-men-tablebases

Author: Angrim

Date: 13:22:45 09/08/02

Go up one level in this thread


On September 08, 2002 at 07:33:54, GuyHaworth wrote:
>
>The practice of generating 6-man EGTs not only generates the EGTs but generates:
>
>- experience in generating EGTs
>- awareness of the problem that distribution of the information is not easy
>
>New ideas are coming forward from time to time, and these will be even more
>useful for larger EGTs.
>
>e.g.:  If we had a GIMPS- or SETI-like distributed-community approach to
>generating EGTs, an HQ distributing the remaining unsolved problems, we would
>not be taking years to generate 6-man EGTs.
>
>We are not using parallel computers ... or distributed communities yet.

Parallel computers were used for this in the distance past.  Roughly
a week after fire was invented someone used a cray to build a set of
6man egtb *in ram* and published the statistics about each table.
As far as I know these tables were never written to disk/tape though.

A SETI-like distributed-community approach only works for problems that
are mostly computation with little need for ram/disk space.  There is
no hope of it being used for egtb generation.

Of course, a few people with access to really nice computers and
really fast network connections could split the work up, but this
wouldn't be any better than a single person with access to a bunch of
really nice computers(like say.. someone working at MS in the
compiler division) building on multiple computers.

>Nor are we generating DTC or DTZ EGTs (which involve smaller depths and
>therefore less space), or getting a better compression in 'real-time versions of
>the EGTs' by setting 'broken' values' to 'last seen' values.
>
>This last is admittedly a 'finesse' and assumes that no 'broken' positions are
>enquired about - or at least that the access-code checks for 'broken positions'.

There are certainly a number of ways to make the data more compressable.
Also the data compression method could be improved.
Some of these changes (ie DTM->DTC) would require that everyone
using the tables make changes to their programs, and thus are unlikely.
Others such as setting "broken" values to something meaningless but
easy to compress would not require work by engine authors, and so might
happen someday.

>
>g

Angrim



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.