Stockfish github has nice syzygy test
Posted: Thu Jan 16, 2020 9:32 am
2-6 pieces and up to 80 000 games. But I don't find it anymore. Where is it?
Page 1 of 1
Re: Stockfish github has nice syzygy test
Posted: Thu Jan 16, 2020 12:01 pm
Of course I mean 3-6 piece. And 3-4 gave nothing as SF has basic knowledge.
Re: Stockfish github has nice syzygy test
Posted: Thu Jan 23, 2020 9:15 am
It was cleverly hidden https://github.com/glinscott/fishtest/wiki/UsefulData.
Elo gain using syzygy
Tested at 10+0.1, with all syzygy WDL files on tmpfs (i.e. RAM), testing using none(0), 4, 5, and 6 man TB in a round-robin tournament (SF10dev).
Rank Name Elo +/- Games Score Draws
1 syzygy6 13 2 82591 51.8% 59.5%
2 syzygy5 2 2 82590 50.3% 59.4%
3 syzygy4 -7 2 82591 49.0% 59.3%
4 syzygy0 -7 2 82592 48.9% 59.4%
Tested at 60+0.6, with all syzygy WDL files on tmpfs (i.e. RAM), testing using none(0) against 6 man TB:
Score of syzygy6 vs syzygy0: 4084 - 3298 - 18510 [0.515] 25892 Elo difference: 10.55 +/- 2.25
So 6 piece = 149 GB data gives STC 20 ELO and LTC 10 ELO. Maybe at tournament level and SMP 5 ELO only
! Waste of storage?
Elo gain using syzygy
Tested at 10+0.1, with all syzygy WDL files on tmpfs (i.e. RAM), testing using none(0), 4, 5, and 6 man TB in a round-robin tournament (SF10dev).
Rank Name Elo +/- Games Score Draws
1 syzygy6 13 2 82591 51.8% 59.5%
2 syzygy5 2 2 82590 50.3% 59.4%
3 syzygy4 -7 2 82591 49.0% 59.3%
4 syzygy0 -7 2 82592 48.9% 59.4%
Tested at 60+0.6, with all syzygy WDL files on tmpfs (i.e. RAM), testing using none(0) against 6 man TB:
Score of syzygy6 vs syzygy0: 4084 - 3298 - 18510 [0.515] 25892 Elo difference: 10.55 +/- 2.25
So 6 piece = 149 GB data gives STC 20 ELO and LTC 10 ELO. Maybe at tournament level and SMP 5 ELO only
! Waste of storage?Re: Stockfish github has nice syzygy test
Posted: Thu Jan 23, 2020 12:07 pm
Hmmm, curious scaling. I would expect stronger play would be decided more often in endgames. Thanks for pointing, needs explanation, the increased draw rate here from 59% to 71% cannot stand alone for explanation.Jouni wrote: ↑Thu Jan 23, 2020 9:15 am It was cleverly hidden https://github.com/glinscott/fishtest/wiki/UsefulData.
Elo gain using syzygy
Tested at 10+0.1, with all syzygy WDL files on tmpfs (i.e. RAM), testing using none(0), 4, 5, and 6 man TB in a round-robin tournament (SF10dev).
Rank Name Elo +/- Games Score Draws
1 syzygy6 13 2 82591 51.8% 59.5%
2 syzygy5 2 2 82590 50.3% 59.4%
3 syzygy4 -7 2 82591 49.0% 59.3%
4 syzygy0 -7 2 82592 48.9% 59.4%
Tested at 60+0.6, with all syzygy WDL files on tmpfs (i.e. RAM), testing using none(0) against 6 man TB:
Score of syzygy6 vs syzygy0: 4084 - 3298 - 18510 [0.515] 25892 Elo difference: 10.55 +/- 2.25
So 6 piece = 149 GB data gives STC 20 ELO and LTC 10 ELO. Maybe at tournament level and SMP 5 ELO only
Re: Stockfish github has nice syzygy test
Posted: Thu Jan 23, 2020 5:54 pm
149GB is a fly spitting into the wind. Even if the occasional extra win was not a whole ELO point it would be worth it. Now if full 7 piece or 8, gave less than 5 ELO a debate about wasted space could be had.
At current storage costs 149GB is imperceptible bit dust. I have a thumb drive with more storage plugged into my TV.
At current storage costs 149GB is imperceptible bit dust. I have a thumb drive with more storage plugged into my TV.
Re: Stockfish github has nice syzygy test
Posted: Thu Jan 23, 2020 7:07 pm
I agree with the numbers 5 to 20 Elo depending on the time controls. And this is what I see in my testing. But this is not the whole story. What does this Elo gain mean. It can mean 1 to 3 extra wins or saved losses in every 100 games. And Stockfish uses TB's very well because of the SF search, NN on the other had do not IMO. If I throw out 149 GB of Table Bases to save storage, on a 2 TB drive. Is this a wise move? No!Jouni wrote: ↑Thu Jan 23, 2020 9:15 am It was cleverly hidden https://github.com/glinscott/fishtest/wiki/UsefulData.
Elo gain using syzygy
Tested at 10+0.1, with all syzygy WDL files on tmpfs (i.e. RAM), testing using none(0), 4, 5, and 6 man TB in a round-robin tournament (SF10dev).
Rank Name Elo +/- Games Score Draws
1 syzygy6 13 2 82591 51.8% 59.5%
2 syzygy5 2 2 82590 50.3% 59.4%
3 syzygy4 -7 2 82591 49.0% 59.3%
4 syzygy0 -7 2 82592 48.9% 59.4%
Tested at 60+0.6, with all syzygy WDL files on tmpfs (i.e. RAM), testing using none(0) against 6 man TB:
Score of syzygy6 vs syzygy0: 4084 - 3298 - 18510 [0.515] 25892 Elo difference: 10.55 +/- 2.25
So 6 piece = 149 GB data gives STC 20 ELO and LTC 10 ELO. Maybe at tournament level and SMP 5 ELO only
Accumulation of small advantages is how you win at chess at this Level!
Re: Stockfish github has nice syzygy test
Posted: Fri Jan 24, 2020 1:09 am
Exactly. I also agree that LCO's tablebase usage is a cruel joke.If I throw out 149 GB of Table Bases to save storage, on a 2 TB drive. Is this a wise move? No!
Accumulation of small advantages is how you win at chess at this Level!