huge opening file like 8moves_GM.pgn?

[BubbaTough]

Hi,

Is there somewhere I can download a file with a lot of starting positions for computer chess testing? I got the impression stockfish was using 8moves_GM.pgn but I have failed to find a copy of it (or something similar).

-Sam

[Adam Hair]

Hi Sam,

This should be the link to where I uploaded the 8moves_GM file:

http://www.mediafire.com/?qb8bt8k9acdq1ko

I think that the Stockfish testers have found a few dubious openings that slipped through my filtering, but it seems to be okay overall. One improvement would be removing the openings that are too drawish.

Adam

[BubbaTough]

Hi Sam,

This should be the link to where I uploaded the 8moves_GM file:

http://www.mediafire.com/?qb8bt8k9acdq1ko

I think that the Stockfish testers have found a few dubious openings that slipped through my filtering, but it seems to be okay overall. One improvement would be removing the openings that are too drawish.

Adam

Thanks Adam!

-Sam

[Adam Hair]

Dariusz Orzechowski and Lucas Braesch found bad openings in 8moves_GM.pgn, and they independently worked on the pgn to remove those positions. I recommend anyone using 8moves_GM.pgn to read the thread at the Fishcooking forum (link given) and download the file that is most appropriate for you:

https://groups.google.com/forum/m/?from ... 8k5kE5yAhM

[Adam Hair]

In case anybody wants the results of Lucas' filtering of 8moves_GM in PGN format, here is a link:

http://www.mediafire.com/download/wvvqb ... _LB.pgn.7z

44,624 openings, all 8 moves (16 plies) deep.

[Graham Banks]

In case anybody wants the results of Lucas' filtering of 8moves_GM in PGN format, here is a link:

http://www.mediafire.com/download/wvvqb ... _LB.pgn.7z

44,624 openings, all 8 moves (16 plies) deep.

Thanks Adam, Dariusz and Lucas. Will take a look.

[beachknight]

In case anybody wants the results of Lucas' filtering of 8moves_GM in PGN format, here is a link:

http://www.mediafire.com/download/wvvqb ... _LB.pgn.7z

44,624 openings, all 8 moves (16 plies) deep.

Thanks Adam. I was looking for such a file for my tests.

[Modern Times]

There is also Kirill's opening sampler:

http://kirill-kryukov.com/chess/tools/opening-sampler/

You can generate a random selection from various opening pgn suites, or indeed grab the entire pgns.

[lucasart]

In case anybody wants the results of Lucas' filtering of 8moves_GM in PGN format, here is a link:

http://www.mediafire.com/download/wvvqb ... _LB.pgn.7z

44,624 openings, all 8 moves (16 plies) deep.

Thanks Adam, Dariusz and Lucas. Will take a look.

Dariusz filter was to remove all positions where a quick SF search gives a score outside +/-50cp.

My filter was more tolerant, because I do not believe in SF's large and volatile scores, especially in the early opening. So I removed openings that showed a score outside +/-50cp according to both SF and Critter (ie. only remove when both SF and Critter agree it's bad).

Anyway, use whichever you want, they are both good. I prefer FEN format personally, because it's much easier to deal with (using linux shell programs, text editors etc.)

[lucasart]

Actually, I'm thinking of creating my own opening book, fram scratch, not using a single PGN from GM games or whatever. It's very simple:
* implement Multi-PV in DiscoCheck
* modify DiscoCheck to read a FEN file, and for each FEN, do a multi-PV (say N PV) search and output the FEN after playing the first move of each PV.
* apply the program iteratively: first pass with a 1 line file of the opening position, gives a output file of N lines. And loop again until the desired depth is reached.
* in the end do a clean up, just as I did with Adam's positions (on the termnal leaves positions).

Voila! The beauty is that it would not suffer the human biais towards simple and solid positions, nor the human creativity limitation. Really a pure search based book. The above could easily be coded in a Python script, so that the same algorithm can be used with any engine supporting multi-PV really.