Crystal 3.0

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Dann Corbit
Posts: 12232
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Crystal 3.0

Post by Dann Corbit » Tue Nov 24, 2020 9:21 am

Andrew wrote:
Tue Nov 24, 2020 7:54 am

I've noticed with many puzzle/study positions, Crystal, some versions of SF and others are able to get a TB mate
score quickly. Then things vary, sometimes the mate score comes soon after sometimes never.

Seems to depend a lot on the endgame, for example sometimes I've seen the end of the PV be KBN endgames
and take a very long time to resolve. Maybe having DTM syzygy bases up to 4 piece could help with this?

Andrew
It would definitely help a lot to have fast 6 or 7 man DTM table base files.
Part of the problem here is that the Syzygy tablebase files do not have distance to mate. So the engine knows what moves win, but it does not know when.

It is not unusual for an engine that cannot find a mate score above 30000 to find a mate if you analyze with the tablebase files turned off because it is forced to search and it gets real scores for the moves. Even so, the tablebase files will often find a win where no tablebase files will miss it

That is why I keep the full 7 man nalimov files on SSD, so that I can find the pretty mate results. I would prefer Syzygy files, but I think the DTM project has dried up, or at least I have heard of no progress in a long time
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Vinvin
Posts: 4998
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Re: Crystal 3.0

Post by Vinvin » Tue Nov 24, 2020 11:40 am

Dann Corbit wrote:
Tue Nov 24, 2020 9:21 am
Andrew wrote:
Tue Nov 24, 2020 7:54 am

I've noticed with many puzzle/study positions, Crystal, some versions of SF and others are able to get a TB mate
score quickly. Then things vary, sometimes the mate score comes soon after sometimes never.

Seems to depend a lot on the endgame, for example sometimes I've seen the end of the PV be KBN endgames
and take a very long time to resolve. Maybe having DTM syzygy bases up to 4 piece could help with this?

Andrew
It would definitely help a lot to have fast 6 or 7 man DTM table base files.
Part of the problem here is that the Syzygy tablebase files do not have distance to mate. So the engine knows what moves win, but it does not know when.

It is not unusual for an engine that cannot find a mate score above 30000 to find a mate if you analyze with the tablebase files turned off because it is forced to search and it gets real scores for the moves. Even so, the tablebase files will often find a win where no tablebase files will miss it

That is why I keep the full 7 man nalimov files on SSD, so that I can find the pretty mate results. I would prefer Syzygy files, but I think the DTM project has dried up, or at least I have heard of no progress in a long time
One idea is to have an option : "Search for a mate when high score (>80.00)" and 2 sub options : "1) Use Syzygy" and "2) don't access Syzygy"
I know this will sometime take a lot of time to access Syzygy in hundreds of variations but it's OK for me. What would you expect from a very high score ? Only a mate score !

The idea is to get correct DTM from this page http://talkchess.com/forum3/viewtopic.p ... 0&p=874099 with less manual operations than before.

Dann Corbit
Posts: 12232
Joined: Wed Mar 08, 2006 7:57 pm
Location: Redmond, WA USA
Contact:

Re: Crystal 3.0

Post by Dann Corbit » Tue Nov 24, 2020 5:11 pm

That would probably help sometimes.
But a single tablebase lookup would be a lot better and also provably minimal searches can be provided (depending on if the TB also understands e.p. and castle rights)
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.

Vinvin
Posts: 4998
Joined: Thu Mar 09, 2006 8:40 am
Full name: Vincent Lejeune

Re: Crystal 3.0

Post by Vinvin » Tue Dec 22, 2020 2:13 am

Dann Corbit wrote:
Tue Nov 24, 2020 5:11 pm
That would probably help sometimes.
But a single tablebase lookup would be a lot better and also provably minimal searches can be provided (depending on if the TB also understands e.p. and castle rights)
One better idea :
- when the best score is not a mate then access only the DTZ files (to get an upper value of the distance to mate)
- when the best score is a mate then access only the WDL files (to find the shortest mate)

User avatar
MikeB
Posts: 4840
Joined: Thu Mar 09, 2006 5:34 am
Location: Pen Argyl, Pennsylvania

Re: Crystal 3.0

Post by MikeB » Tue Dec 22, 2020 3:28 am

jhellis3 wrote:
Thu Sep 10, 2020 6:11 pm
Decided it was an appropriate time for release.

Changes since 2.0:
Various bugfixes mostly due to members of the community pointing out usage issues to me.
Some minor eval smoothing (simple linear interpolation between previous value).
Introduction of NNUE. For now, NNUE is either on or off, there is no hybrid method involved or available.

https://github.com/jhellis3/Stockfish/releases/tag/v3.0
+1 Thanks Joe , much appreciated!
Image

Post Reply