Performance of Syzygy and Scorpio

[Daniel Shawul]

Blah blah ... This is done on your request and found embarrassingly wanting

3 SF 04.02 NO TB : 422.5/1000
4 SF 04.02 Syzygy WDL : 419.0/1000

Only in Syzygy la la land does a NO TB version performs better than one with TB. And now you have the courage to lecture me on how to make progress?? You should have listened to me when I warned you many times. It is not a sin to make a mistake once you know, just admit and fix it.

[Daniel Shawul]

Syzygy WDL doesn't help SF at all.

That's a bit surprising as SF in such positions only probes WDL after a capture or pawn move. The probes after a capture into a 4-piece position probably won't help compared to default SF (since default SF most likely evaluates those 4-piece positions perfectly anyway), but those after pawn moves should be of benefit.

The WDL tables also prevent SF from looking at suboptimal moves and that should help as well.

What kind of positions are you testing on? With pawns or without?

Many do have pawns, generally, from KBNK and KBNKN to KRPKR and such. I could upload the EPD file of hard wins if you need it.

Kai, I don't know if you are working on Stockfish with EGBBs but if you haven't I can make compiles for you. I am sure it will perform much better than Toga.

New 5-men egbbs (211mb): http://shawul.olympuschess.com/egbb/egbb5men.zip
Stockfish source: https://github.com/dshawul/Stockfish/commits/master
Egbbdll source: https://github.com/dshawul/egbbdll/commits/master

[syzygy]

Many do have pawns, generally, from KBNK and KBNKN to KRPKR and such. I could upload the EPD file of hard wins if you need it.

I would be interested in seeing the positions. You could also mail them to syzygy_tb@yahoo.com.

For KBNK the WDL information is of course of no help. For KBNKN there should be a small advantage (SF+WDL will only consider winning moves) but it may be too small. For KRPKR it should help considerably, except that default SF might already evaluate this perfectly.

[syzygy]

And now you have the courage to lecture me on how to make progress??

No, I am just explaining things to the reasonable people that may be reading this thread despite your contributions.

[ouachita]

No, I am just explaining things to the reasonable people that may be reading this thread . . .

As a scientist and novice engine-guy, I sincerely appreciate same

[Adam Hair]

Syzygy WDL doesn't help SF at all.

That's a bit surprising as SF in such positions only probes WDL after a capture or pawn move. The probes after a capture into a 4-piece position probably won't help compared to default SF (since default SF most likely evaluates those 4-piece positions perfectly anyway), but those after pawn moves should be of benefit.

The WDL tables also prevent SF from looking at suboptimal moves and that should help as well.

What kind of positions are you testing on? With pawns or without?

Many do have pawns, generally, from KBNK and KBNKN to KRPKR and such. I could upload the EPD file of hard wins if you need it.

Kai, I don't know if you are working on Stockfish with EGBBs but if you haven't I can make compiles for you. I am sure it will perform much better than Toga.

New 5-men egbbs (211mb): http://shawul.olympuschess.com/egbb/egbb5men.zip
Stockfish source: https://github.com/dshawul/Stockfish/commits/master
Egbbdll source: https://github.com/dshawul/egbbdll/commits/master

I downloaded the source yesterday and compiled it. I have two questions:

1) The default value for 'EgbbProbeDepth' is 75. Should that value be lower?
2) The internal name is Stockfish 040214, but the benchmark matched the benchmark for Stockfish 270114. Which Stockfish version did you use?

[Daniel Shawul]

Syzygy WDL doesn't help SF at all.

That's a bit surprising as SF in such positions only probes WDL after a capture or pawn move. The probes after a capture into a 4-piece position probably won't help compared to default SF (since default SF most likely evaluates those 4-piece positions perfectly anyway), but those after pawn moves should be of benefit.

The WDL tables also prevent SF from looking at suboptimal moves and that should help as well.

What kind of positions are you testing on? With pawns or without?

Many do have pawns, generally, from KBNK and KBNKN to KRPKR and such. I could upload the EPD file of hard wins if you need it.

Kai, I don't know if you are working on Stockfish with EGBBs but if you haven't I can make compiles for you. I am sure it will perform much better than Toga.

New 5-men egbbs (211mb): http://shawul.olympuschess.com/egbb/egbb5men.zip
Stockfish source: https://github.com/dshawul/Stockfish/commits/master
Egbbdll source: https://github.com/dshawul/egbbdll/commits/master

I downloaded the source yesterday and compiled it. I have two questions:

1) The default value for 'EgbbProbeDepth' is 75. Should that value be lower?
2) The internal name is Stockfish 040214, but the benchmark matched the benchmark for Stockfish 270114. Which Stockfish version did you use?

1) 'EgbbProbeDepth' is a bad name for what I am thinking to be ' Percentage of Search depth dedicated for skipping EGBB probes so that it makes better progress'. With 75 it means, 3/4 for the depth will be used to make progress, find mates etc. Maybe there is a better way in UCI to specify percentages

2) The Stockfish version I used is an unofficial one dated to the time I forked it. I don't know if the number changes whenever I compile, but there should be any bugs because I didn't touch anything that doesn't concern EGBBs.

Question for anyone: Why does Arena 3.0 not display tbhits? I send the tbhits along with info pv move ... etc but nothing appears.

[Laskos]

Syzygy WDL doesn't help SF at all.

That's a bit surprising as SF in such positions only probes WDL after a capture or pawn move. The probes after a capture into a 4-piece position probably won't help compared to default SF (since default SF most likely evaluates those 4-piece positions perfectly anyway), but those after pawn moves should be of benefit.

The WDL tables also prevent SF from looking at suboptimal moves and that should help as well.

What kind of positions are you testing on? With pawns or without?

Many do have pawns, generally, from KBNK and KBNKN to KRPKR and such. I could upload the EPD file of hard wins if you need it.

Kai, I don't know if you are working on Stockfish with EGBBs but if you haven't I can make compiles for you. I am sure it will perform much better than Toga.

New 5-men egbbs (211mb): http://shawul.olympuschess.com/egbb/egbb5men.zip
Stockfish source: https://github.com/dshawul/Stockfish/commits/master
Egbbdll source: https://github.com/dshawul/egbbdll/commits/master

Daniel, can you make compiles (both SF and dll) while I am downloading the new 5-men? I don't want to mess with sources.

[Daniel Shawul]

I compiled an x64-windows binary for Stockfish and Scorpio as well.

https://dl.dropboxusercontent.com/u/552 ... ngines.zip

Two egbbdlls are also included, so you can move those to the egbb directory and replace the ones you dowloaded.
It doesn't really matter much for the test but it avoids some memory leaks. I can compile a 32 binary if you need it.

[Laskos]

I compiled an x64-windows binary for Stockfish and Scorpio as well.

https://dl.dropboxusercontent.com/u/552 ... ngines.zip

Two egbbdlls are also included, so you can move those to the egbb directory and replace the ones you dowloaded.
It doesn't really matter much for the test but it avoids some memory leaks. I can compile a 32 binary if you need it.

SF hangs, after asking for some dll-s from Visual C++ Redistributable for Visual Studio 2012 Update 4, which I installed.