Eval tuning data

[gaard]

I've gained approximately 100 Elo tuning with E12.41-1M-D12-Resolved.book by Andrew Grant, but I'm not sure I can go any further with it. Can anyone suggest something comparable that I can tune my eval with? I vaguely remember others advocating similar datasets but I can't readily find them. Thanks in advance.

[op12no2]

Assuming you mean hce tuning...

Rebel has some here, but keep in mind the wdl result is relative to stm:-

http://rebel13.nl/download/data.html

Alexandru Moșoi's (Zurichess) quiet_labeled.epd seems to give good results for many:-

https://bitbucket.org/zurichess/tuner/downloads

Have you considered generating your own data? Easy to get the PGNs with cutechess etc. and there are some existing tools around to convert to epd+wdl. Then you need to filter 'quiet' positions to your liking with a tweak to your engine or a script using your engine. Personally I find it psychologically more satisfying than using third party data.

If you are a Windows user a fast pgn to epd+wdl tool is called pgn-extract.

https://www.cs.kent.ac.uk/people/staff/djb/pgn-extract

e.g.

.\pgn-extract.exe -Wepd --nofauxep -s -o lozza.epd lozza.pgn

[JVMerlino]

I second using Alexandru's epd file. I got excellent results with it.

But I can't find Andrew Grant's file that the OP refers to. Where can it be downloaded?

[op12no2]

I second using Alexandru's epd file. I got excellent results with it.

But I can't find Andrew Grant's file that the OP refers to. Where can it be downloaded?

http://www.talkchess.com/forum3/viewtop ... =7&t=75350

(I think)

[JVMerlino]

Thank you!

[AndrewGrant]

Reminder that those datasets are mildly mislabeled, and one of the standard ones if FRC and the FRC is standard. So pay attention, just cat out the top of the file and it will be obvious.

[jhonnold]

Berserk had a lot of luck with this dataset.

This is a pre-filtered combination of the datasets mentioned in this thread.