Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Problem with P.Hash and different contempt settings?
- - By Martin Hander (*) Date 2008-08-11 10:16 Edited 2008-08-11 10:32
After playing around with Rybka3 for a while, playing games and making analysis,
I realized that I always used the the same Persistent Hash File.

Doesn't that mean, that my Persistent Hash File is now filled with entries
with different contempt settings (0, 15 and sometimes 100),
which makes it completely useless for every purpose?

Rybka should only use the hash entries which are calculated with the same contempt value
as the actual settings, but this is only possible if the contempt setting is saved for every hash entry.

Perhaps someone knows if this is the case, I found nothing in the FAQ.

Another possible solution:
A Persistent Hash File can be used only for one contempt setting (contempt value in the file name or header).
Parent - By Uly (Gold) Date 2008-08-11 18:56

> playing games


I don't know about your other questions but for games I'd switch PH off.

For analysis, the hash should fix itself once you reach a new ply, as I put 200 cpawns contempt for a game against Crafty, then it was analyzing the starting position as -1.87 for black after I reset contempt, but it fixed as soon as the next ply appeared.
Parent - By Vasik Rajlich (Silver) Date 2008-08-11 20:52
Good question and good point. Rybka stores entries in the persistent hash without worrying about the contempt. It's up to users to take this into account.

I'll add this to the FAQ.

Vas
Parent - - By cma6 (****) Date 2008-08-12 02:35 Edited 2008-08-12 02:40
I have been using Rybka 3 with Persistent Hash turned off (as well as Preserve Analysis turned off) resulting in  faster, better analysis. I think Persistent Hash is a possible detriment to analysis, except in the unusual situation where you expect to resume analysis of the same game at a later session, in which case PH is useful.

  There is one aspect for PH which I would love to be able to use, though I don't know if it can be done. When you reach an evaluation of a position, say after move 23, with which you are content, it would be great to be able to propagate backwards that evaluation to, say, move 20, and begin analysis of another branch with that later evaluation being used.

  It seems to me that that would be the real value of PH if it were possible. (I know that Preserve Analysis does that, but the hit to performance is too great with PA.)
Parent - By Vasik Rajlich (Silver) Date 2008-08-12 17:38

>  I think Persistent Hash is a possible detriment to analysis, except in the unusual situation where you expect to resume analysis of the same game at a later session, in which case PH is useful


PH should have no negative side-effects.

> There is one aspect for PH which I would love to be able to use, though I don't know if it can be done. When you reach an evaluation of a position, say after move 23, with which you are content, it would be great to be able to propagate backwards that evaluation to, say, move 20, and begin analysis of another branch with that later evaluation being used


That's the entire point of PH.

Vas
Up Topic Rybka Support & Discussion / Rybka Discussion / Problem with P.Hash and different contempt settings?

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill