Stockfish 3 PA_GTB

[Ace1]

3-fold rep not working properly..
In some cases its ok

[Ace1]

3-fold rep not working properly..
In some cases its ok

Sorry
Its working

[endings]

my mistake

[Luigi335]

Thank you, Jeremy.

The fourth version of Stockfish 3 PA GTB perfectly works on my i7 2630 QM (4 physical cores) with ChessBase GUI.

It is certainly the best SF for Gaviota TB.

Were it possible to have also a version that supports Nalimov TB6?

[Jeremy Bernstein]

Thank you, Jeremy.

The fourth version of Stockfish 3 PA GTB perfectly works on my i7 2630 QM (4 physical cores) with ChessBase GUI.

It is certainly the best SF for Gaviota TB.

Were it possible to have also a version that supports Nalimov TB6?

Nope. Nalimov probing code requires a use permission which I'm not interested in trying to get for a hobby project.

Sorry,
Jeremy

[Vael Jean-Paul]

Hi Jeremy,

Thank you for this 004 version..working great here..!
GTB's showing TB-hits much better then 003 and for some reason it plays much stronger then 003!

What do you think about using the RTB's from Ronald de Man has now 6-piece tables with a very small sized files:
The 6-piece WDL tables are 68.2 GB in total. The DTZ tables take up 81.9 GB. For up to 5 pieces, the numbers are 378 MB and 561 MB. Ideally, the WDL tables are stored on an SSD.
https://github.com/syzygy1/tb

Is it possible later to try using these RTB's in a Stockfish version!?

Kind regards,
Jean-Paul.

[Jeremy Bernstein]

Hi Jeremy,

Thank you for this 004 version..working great here..!
GTB's showing TB-hits much better then 003 and for some reason it plays much stronger then 003!

What do you think about using the RTB's from Ronald de Man has now 6-piece tables with a very small sized files:
The 6-piece WDL tables are 68.2 GB in total. The DTZ tables take up 81.9 GB. For up to 5 pieces, the numbers are 378 MB and 561 MB. Ideally, the WDL tables are stored on an SSD.
https://github.com/syzygy1/tb

Is it possible later to try using these RTB's in a Stockfish version!?

Kind regards,
Jean-Paul.

Great to hear, thanks. The rewritten probing criteria probably help somewhat, and the inclusion of 3x repetition under certain circumstances, as well. In version 005, 3x repetition detection will be enabled all the time; we'll see what kind of impact that has. In testing, it eliminates a bunch of false 0.00 evaluations which cause SF to make some questionable choices for a negligable cost in time.

Ronald de Man has his own version of Stockfish which has been modified to take advantage of his TBs. When I did the rewrite for version 3, I even used some of his probing logic as a basis for the GTB probes (although I've removed/rewritten most of it in the meantime for a variety of reasons). I would consider trying to support the RTBs as an option in the future, but probably not anytime real soon.

My first priority is a caveman approach to position learning/persistent hash (which is nearly ready for release, in fact, and working pretty well), followed by some explorations into how to improve Stockfish's ability to analyze certain positions with very deep tactical lines beginning with quiet moves.

I'm looking at one (which I can't show you yet as it's part of someone's ongoing game) in which the best move is a quiet move which allows a check. It gets quickly pushed down to position 5 or 6 in the move list and never gets a chance for a full exploration. There may be no solution for that, though -- even after forcing the move, SF doesn't recognize its superiority until it reaches depths beyond d36). According to one tester, SFs v1.7 and 1.8 found the move, with the v1.9 changes causing it to get ignored. Anyway, that's what I'm looking into right now.

Best,
Jeremy

[Luigi335]

The manual control of "Soft Probe Depth" and "Hard Probe Depth" is a very good idea!

An interesting setting for these options (and for long time controls) is:

Soft Probe Depth = 16
and
Hard Probe Depth = 24

that are the Houdini's default values, and it seems very good.

On my GUI I have saved three different sets of values: 10/16 (the Jeremy's default), 16/24 (the Houdart's default) and 16/32, and I am testing them on a set of end game positions.

I haven't a Solid State Disk for TB and I think that is very important don't excessively slow down the speed of the engine, for not penalizing the depth of the search.

[Jeremy Bernstein]

v005 is here!

New features:
- Persistent Hash implementation (see included file "About Persistent Hash.txt" for more information)
- non-main threads can publish their currmovenumber
- 3x repetition is checked all the time
- update to the tip of the Stockfish upstream master branch

Thanks to my diligent and dedicated testers for the great work on getting the Persistent Hash feature working usefully and reliably.

Jeremy

[User923005]

Thanks for your work on this. It is very useful for me to have a stockfish that probes the endgame. I use chess engines primarily for analysis rather than game play. Hence, I need the right answer, and I am willing to wait to get it if needed. Deep analysis (e.g. several hours per position) will often start to hit the tablebases with a surprising number of chessmen on the board, initially.

I think that people buy chess engines to win online games, for the most part. But if you want to become a better chess player, analysis is far more important than game play, which does not teach you very much, if we are real about it.

For this reason, tablebase files are important, because the chess engine will make the best possible move if given enough time. We have no such guarantee if the tablebase files are missing.

I think that analysis mode (e.g. when we send the directive to analyze) should probably use a bit different search and evaluation compared to game play. For instance, in game play, underpromotion usually hurts speed, especially underpromotion to rook or bishop. But if we definitely want the best answer and not to 'win a game in a hurry' then underpromotion is required. So, perhaps, for game play an engine could not underpromote, but for analysis it would.

Just a thought.

Thanks again for your diligence and service in providing this fine tool. And what is wrong with the stockfish team to not incorporate tablebase probes into their product? For crying out loud.