Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: EGTB algoritmic question

Author: Simon Finn

Date: 05:45:09 07/11/02

Go up one level in this thread


On July 10, 2002 at 16:23:54, Daniel Clausen wrote:

>On July 10, 2002 at 14:17:04, GuyHaworth wrote:
>
>>
>>Yes, if you are computing your EGT(B) to the DTM(ate) metric.
>>
>>You might find a 'mate in many moves', only to discover that there was a much
>>quicker 'mate in much few moves' which appears many retrograde_cycles later in
>>the computation.
>>
>>For example, I forget the exact position but you will get the idea:
>>
>>wQb3wBh1h2/bKa1bNe1+w ....
>>
>>1.Qa2+ forcing Kxb2 leaves (I guess) a deep KBBKN win
>>
>>This mate would be found two cycles into the retrograde_analysis of KQBBKN but
>>a much quicker win using the Queen would turn up later.
>
>Well yes, but I still don't see a problem. If you only mark the positions which
>are 'mate in 2' during the 2nd cycle and leave the other positions untouched,
>you don't have this problem. The best you could do for this position in the 2nd
>cycle is to mark it as a 'potentially mate in 1+(DTM-value of KBBKN)' [a mate in
>X in iteration Y is only a potential mate if X>Y]
>
>I don't see a reason why I would want to do this, especially if this 'error'
>even propagates to other positions during later iterations. So why not simply
>leave the position unmarked in this case? Would generating the table be that
>more slow?

The approach you describe requires max-DTM iterations. For some
endgames - such as KBNKR (where all wins are based on a quick
mate or quick conversion) - this is substantially more than
with the alternative technique.

You also need to keep the conversion databases available
throughout the algorithm, rather than being able to
discard them after a pre-pass. This can increase memory
usage and/or disk I/O.

Simon


>
>Sargon



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.