Syzygy Tablebase and a Request to Jeremy...

[Luigi335]

Syzygybases is a new promising tablebase format developed by Ronald de Man (nick name "Syzygy).

Syzygybases take 50 moves rule into account, the probing code is distributed under permissive license, and the table are about 7 - 8 times smaller than Nalimov's tablebases (1 GB for TB5 and 150 GB for TB6).

I think this could become the standard tablebase format for the next decade of computer chess.

The TB can be generated, see https://github.com/syzygy1/tb

but this requires a very powerful hardware, or downloaded by: http://oics.olympuschess.com/tracker/index.php

At the moment, there are some versions of Stockfish that support this format, for example: http://rghost.net/48928598

After a long wait of about 10 days (unfortunately in Italy Internet speed is that of a Third World country...), at last, I have downloaded the 150 GB of Syzygy 6men DTZ & WDL.

With the 5men, SF perfectly works and shows very high speed of access to the TB, really better than Gaviota (and Nalimov for other engines).

Nevertheless, with Syzygy 6men, Stockfish has the same problem that afflicts other engines with Nalimov 6men: an excessive slowdown of the engine, that spends much time in the search of data base positions rather than in the analysis.

It would be welcome a version of SF Syzygy bases with "Soft Probe Depth" and "Hard Probe Dept" adjustment, as in Houdini and in Jeremy's SF3 PA GBT...

[Jeremy Bernstein]

Hi there,

It's going to take a while, but I'm open to exploring that -- I've been following the Syzygy developments with some interest. But I don't have so much time for playing around with chess engines at the moment, I'm afraid.

jb

[Luigi335]

Ronald de Man has added a new option, "SyzygyProbeDepth" to the latest versions of Stockfish.

http://abrok.eu/stockfish_syzygy/

In this way I was finally able to use 6 man TB without appreciable slowing of the engine.

The default value is "1", which corresponds to the older behaviour. But I set it to a higher value of "20". In this way, the access to TB on my slow hard disk are very small, but very effective, and there are no leaks in the depth of analysis.

I think that so the TB come into play right when they are needed, that is when on the board there are only a dozen pieces or less.

[Jeremy Bernstein]

I finally have a set of syzygy bases built -- only took about 8 days on my 4-core machine. I might look into adding them as an option to the PA_GTB version, but I feel like there are numerous folks working on various syzygy/SF implementations out there. OTOH, if PA_GTB is about analysis, I guess supporting 6-man bases is an obvious improvement. I wouldn't remove GTB support, though. Anyway, none of that is going to happen immediately.

In any case, it sounds like you got what you wanted in the official RdM version.

[Luigi335]

Yes, Jeremy, GTB support is important, and also Robert Houdart has Kept it in his Houdini 4.

Moreover, Ronald de Man was forced to add, in the last versions, an option to "remove the 50-move rule", because it seems that the International Correspondence Chess Federation wants to delete or change this rule...

[Jeremy Bernstein]

Yes, Jeremy, GTB support is important, and also Robert Houdart has Kept it in his Houdini 4.

Moreover, Ronald de Man was forced to add, in the last versions, an option to "remove the 50-move rule", because it seems that the International Correspondence Chess Federation wants to delete or change this rule...

Although I am not a big fan of modern corr chess, I would consider adding a similar option to disable the 50-move rule (or just using RdM's patch) if people are actively using SF PA_GTB for that. I certainly wouldn't want anyone to miss out on a TB mate in 167.