Problem with 6-piece syzygy-bases using wine

Discussion of chess software programming and technical issues.

Moderator: Ras

syzygy
Posts: 5774
Joined: Tue Feb 28, 2012 11:56 pm

Re: Problem with 6-piece syzygy-bases using wine

Post by syzygy »

BBauer wrote:No, it doesn't work with regular Stockfish either.
It's broken.
If multipv is broken, you might want to report that to the SF team (if you haven't done so already).
When analyzing games Stockfish will load the table bases at each move.
I'm not sure what you mean by this. Is there a problem with the tablebases, or do they work as they should?
But we have some elo gain.
Of course, it's not your fault.
I just had a look at the SF source. Verification searches are still performed, but at higher depth than some time ago.
BBauer
Posts: 658
Joined: Wed Mar 08, 2006 8:58 pm

Re: Problem with 6-piece syzygy-bases using wine

Post by BBauer »

You wrote:
I'm not sure what you mean by this. Is there a problem with the tablebases, or do they work as they should?

No, there is no problem with the table bases, but they are initiallized each time I do a move. Before they were initialized only on start. That means I cannnot analyze a set of positions in arena.

As I compile Stockfish myself, do you think I can lower the verification parameter?

I have not reported the problem to the Stockfish team, simply because I am not aware how to do. (Stupid me)

Kind regards
Bernhard
syzygy
Posts: 5774
Joined: Tue Feb 28, 2012 11:56 pm

Re: Problem with 6-piece syzygy-bases using wine

Post by syzygy »

BBauer wrote:You wrote:
I'm not sure what you mean by this. Is there a problem with the tablebases, or do they work as they should?

No, there is no problem with the table bases, but they are initiallized each time I do a move. Before they were initialized only on start. That means I cannnot analyze a set of positions in arena.
Ah, I see what you mean. The same happened for me. For some reason Arena keeps restarting the engine. I agree this is very annoying.

Does Arena not do this if you use the Windows version of SF instead of the Linux version?
As I compile Stockfish myself, do you think I can lower the verification parameter?
Yes, I think the line

Code: Select all

  if (depth < 12 * ONE_PLY)
used to be

Code: Select all

  if (depth < 6 * ONE_PLY)
but there may well have been other changes as well. I don't know if going back to 6 will result in SF finding the win in the position you posted.
BBauer
Posts: 658
Joined: Wed Mar 08, 2006 8:58 pm

Re: Problem with 6-piece syzygy-bases using wine

Post by BBauer »

Now, using the latest version I do no longer see the bases loaded each time I make a move.
Reducing depth to lower 6*one_ply does not help. There is a lot of other pruning.
For my purposes sting-4 is a great help, only thing I miss there are the syzygy bases. -:)

Your bases work very well with 4 threads and 5-pieces. Since I have no SSD performance is not so great with 6-pieces.
Anyway

Thank you for your help

Bernhard
syzygy
Posts: 5774
Joined: Tue Feb 28, 2012 11:56 pm

Re: Problem with 6-piece syzygy-bases using wine

Post by syzygy »

BBauer wrote:Now, using the latest version I do no longer see the bases loaded each time I make a move.
You mean the latest version of Stockfish? (Or did you upgrade Arena.)
Your bases work very well with 4 threads and 5-pieces. Since I have no SSD performance is not so great with 6-pieces.
I'll add a parameter to limit probing to a certain (remaining) depth. That should help.
syzygy
Posts: 5774
Joined: Tue Feb 28, 2012 11:56 pm

Re: Problem with 6-piece syzygy-bases using wine

Post by syzygy »

syzygy wrote:
BBauer wrote:Your bases work very well with 4 threads and 5-pieces. Since I have no SSD performance is not so great with 6-pieces.
I'll add a parameter to limit probing to a certain (remaining) depth. That should help.
Done.
phenri
Posts: 284
Joined: Tue Aug 13, 2013 9:44 am

Re: Problem with 6-piece syzygy-bases using wine

Post by phenri »

New option SyzygyProbeDepth.

The tables are only probed at nodes with remaining depth >= SyzygyProbeDepth.
The default value is 1, which corresponds to the old behaviour.
Set it to a higher value if probing affects nps too much.
Hi RDM, thank you for this.
Please, explain the parameters.
If I put 2, how to interpret this setting? and 100?
What is the parameter that would be a good average?
syzygy
Posts: 5774
Joined: Tue Feb 28, 2012 11:56 pm

Re: Problem with 6-piece syzygy-bases using wine

Post by syzygy »

phenri wrote:Please, explain the parameters.
If I put 2, how to interpret this setting? and 100?
What is the parameter that would be a good average?
If you set it at 2, it will not probe in the leaves of the search tree, but it will probe in the parents of the leaves and higher. In general, it will only probe in nodes with remaining depth at least SyzygyProbeDepth.

Practically speaking: the higher you set it, the fewer probes it will make. Try to find a value where it does not slow down the search too much (in terms of nps). This will depend on your system (how much RAM etc.).
phenri
Posts: 284
Joined: Tue Aug 13, 2013 9:44 am

Re: Problem with 6-piece syzygy-bases using wine

Post by phenri »

syzygy wrote:
phenri wrote:Please, explain the parameters.
If I put 2, how to interpret this setting? and 100?
What is the parameter that would be a good average?
If you set it at 2, it will not probe in the leaves of the search tree, but it will probe in the parents of the leaves and higher. In general, it will only probe in nodes with remaining depth at least SyzygyProbeDepth.

Practically speaking: the higher you set it, the fewer probes it will make. Try to find a value where it does not slow down the search too much (in terms of nps). This will depend on your system (how much RAM etc.).
Thank you,
To understand is that the values ​​of [1-100] are linear? Or is [1, 100-2]?
syzygy
Posts: 5774
Joined: Tue Feb 28, 2012 11:56 pm

Re: Problem with 6-piece syzygy-bases using wine

Post by syzygy »

phenri wrote:To understand is that the values ​​of [1-100] are linear? Or is [1, 100-2]?
If you mean linear in the sense of linearly decreasing the number of accesses, then the answer is no. The number of accesses will decrease logarithmically with an increasing (remaining) probe depth.