Page 1 of 1

BPOCG

Posted: Mon Nov 09, 2020 6:34 pm
by Dann Corbit
I have had some requests for my collection of correspondence games, henceforth known as "Big Pile Of Correspondence Games".
It turns out my memory was a little fuzzy. I don't have a million correspondence games, I have two million.
Anyway, this is a link to the 7zip compressed PGN file:
https://drive.google.com/file/d/1lezGRJ ... sp=sharing

Get it fast because I won't leave it up forever.
About 205MB compressed and 1.3GB decompressed.

Re: BPOCG

Posted: Mon Nov 09, 2020 7:26 pm
by Volker Pittlik
Thanks Dann!

Re: BPOCG

Posted: Mon Nov 09, 2020 7:43 pm
by Rebel
Dann Corbit wrote:
Mon Nov 09, 2020 6:34 pm
I have had some requests for my collection of correspondence games, henceforth known as "Big Pile Of Correspondence Games".
It turns out my memory was a little fuzzy. I don't have a million correspondence games, I have two million.
Anyway, this is a link to the 7zip compressed PGN file:
https://drive.google.com/file/d/1lezGRJ ... sp=sharing

Get it fast because I won't leave it up forever.
About 205MB compressed and 1.3GB decompressed.
That's a nice collection, thanks.

It gave 2.2 million new positions, in total now 31 million.

Did a tryout analysis 100,000 positions, SF12, depth=20, 18 threads, time 2h:55m, rounded, 3 hours.
CPU - Intel(R) Xeon(R) CPU E5-2680 v2 @ 2.80GHz

For 31 million positions (18 threads) that would mean 310*3=930 hours = 38 days and 18 hours.

Re: BPOCG

Posted: Mon Nov 09, 2020 7:46 pm
by Dann Corbit
Afterwards, I suggest to mini-max them.

Re: BPOCG

Posted: Mon Nov 09, 2020 9:00 pm
by Dann Corbit
I guess that almost all of them will be analyzed already by Bojun Guo.
Now, a lot of his data will be analyzed with a slightly older version of SF, so you might want to go ahead and analyze for that reason, but I guess almost all of them will already have been analyzed and stored in his dataset.
A lookup will be a lot faster than an analysis pass. If you download his full database, the lookups will be local rather than over the net and hence greasy fast.

Re: BPOCG

Posted: Mon Nov 09, 2020 9:55 pm
by Rebel
Dann Corbit wrote:
Mon Nov 09, 2020 9:00 pm
I guess that almost all of them will be analyzed already by Bojun Guo.
Now, a lot of his data will be analyzed with a slightly older version of SF, so you might want to go ahead and analyze for that reason, but I guess almost all of them will already have been analyzed and stored in his dataset.
A lookup will be a lot faster than an analysis pass. If you download his full database, the lookups will be local rather than over the net and hence greasy fast.
Yes, I forgot about his excellent work.