Lc0 Training Games DBs

This is the repository I keep of Lc0 training game PGNs. These are not the highest quality games, but there sure are a lot of them. Each PGN should have somewhere in the neighborhood of one to two million games. There's one PGN per day, for a test run that started in 2022. But T80 is just one of maybe ten different test runs. There are probably a billion games in that open directory.

https://www.mediafire.com/folder/q5yrdd9q47jzi/Lc0+Training+Games

CERL 240820-1

Rank Name                          Elo     +/-   Games   Score    Draw
   1 Stockfish 1.8                 100       7    5000   64.0%   42.6%
   2 Stockfish 1.7.1                90       7    5000   62.6%   42.8%
   3 Stockfish 1.7                  81       7    5000   61.4%   43.5%
   4 Stockfish 1.6.3                 4       7    5000   50.6%   42.2%
   5 Stockfish 1.5                -136       8    5000   31.3%   36.6%
   6 Stockfish 1.4                -148       8    5000   29.9%   34.3%

Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz
Cores: 8
Logical processors: 16

CuteChess 1.3.1 CLI
10s+0.1 • No book • No adjudication • No tablebases • Ponder off • 5 concurrent
Opening suite: named-openings.pgn to whichever ply is necessary
All 15,000 games

CERL 240819-1

Rank Name                          Elo     +/-   Games   Score    Draw
   1 4ku 4.0                       101       8    5000   64.1%   26.6%
   2 Stockfish 1.5                  11       8    5000   51.6%   34.9%
   3 Stockfish 1.4                   6       8    5000   50.9%   33.9%
   4 Stockfish 1.3                 -27       8    5000   46.2%   34.2%
   5 Stockfish 1.3.1               -30       8    5000   45.6%   36.1%
   6 Stockfish 1.2                 -60       8    5000   41.5%   30.8%

Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz
Cores: 8
Logical processors: 16

CuteChess 1.3.1 CLI
10s+0.1 • No book • No adjudication • No tablebases • Ponder off • 5 concurrent
Opening suite: named-openings.pgn to whichever ply is necessary
All 15,000 games

CERL 240818-1

Rank Name                          Elo     +/-   Games   Score    Draw
   1 4ku 5.0                       138       9    5000   68.9%   24.7%
   2 4ku 4.0                       127       9    5000   67.6%   25.9%
   3 Stockfish 1.2                 -30       8    5000   45.7%   31.4%
   4 Stockfish 1.1                 -54       8    5000   42.3%   30.5%
   5 Stockfish 1.01                -60       8    5000   41.4%   30.1%
   6 Stockfish 1.0                -114       8    5000   34.2%   30.7%

Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz
Cores: 8
Logical processors: 16

CuteChess 1.3.1 CLI
10s+0.1 • No book • No adjudication • No tablebases • Ponder off • 5 concurrent
Opening suite: named-openings.pgn to whichever ply is necessary
All 15,000 games

CERL 240815-1

Rank Name                          Elo     +/-   Games   Score    Draw
   1 4ku 5.0                       182       9    5000   74.1%   29.8%
   2 4ku 4.0                       180       9    5000   73.8%   30.0%
   3 4ku 3.1                        25       8    5000   53.6%   30.8%
   4 4ku 3.0                        17       8    5000   52.5%   30.3%
   5 4ku 2.0                      -134       9    5000   31.7%   24.9%
   6 4ku 1.1                      -310      11    5000   14.4%   16.2%

Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz
Cores: 8
Logical processors: 16

CuteChess 1.3.1 CLI
10s+0.1 • No book • No adjudication • No tablebases • Ponder off • 5 concurrent
Opening suite: named-openings.pgn to whichever ply is necessary
All 15,000 games

Online Strong Games

Strong games are defined by ChessBase as no less than ten moves, with players at least 2300. To filter from either the Chesscom Elite Database files, or from the Lichess Open Database files, I use the following command in Windows PowerShell:

-t"tags.txt" -R"roster.txt" --xroster --minmoves 10 --maxmoves 250 --fixresulttags --fixtagstrings --nosetuptags -C -V -N -D -e --plycount -w9999

This takes the form of: .\pgn-extract INPUT INCLUDED_TEXT -o OUTPUT

For example:
.\pgn-extract original.pgn -t"tags.txt" -R"roster.txt" --xroster --minmoves 10 --maxmoves 250 --fixresulttags --fixtagstrings --nosetuptags -C -V -N -D -e --plycount -w9999 -o processed.pgn

This line takes into account two other text files: tags.txt and roster.txt. This is the full contents of tag.txt:

WhiteElo >= "2300"
BlackElo >= "2300"
TimeControl >= "180"

 

And here are the full contents of the roster.txt file:

Event
Site
Date
Round
White
Black
Result
WhiteElo
BlackElo
ECO
Opening
Variation
EventCountry
EventDate
SetUp
FEN
Variant
TimeControl
Beauty
PlyCount

 

The online strong games files can be found in two directories in the Chess Nerd PGN Repository:
https://chessnerd.net/pgn/chesscom/cc-strong/
https://chessnerd.net/pgn/lichess/lichess-strong/

Complete Named Openings

Here is the complete list of named openings (obtained via Lichess on GitHub) in different formats:
https://www.mediafire.com/file_premium/soqtsohr66rq8wf/openings.zip/file

TSV: This contains the ECO code, the name, the moves in both SAN and UCI, and the FEN.

PGN: This contains a full PGN for each opening. There is no FEN, as that describes the final position, and there is no UCI. Because it’s not in the list, the Double Bongcloud happens to have been added to the PGN only. If you need to use this as an opening suite, all you need do is remove Fool’s Mate, which can be found by searching on the hash symbol (#).

PDF: Made from a DOCX which was made from an HTML file. This includes ECO code, name, moves in SAN and UCI, FEN, and a rendered board of the FEN.

Total size of the ZIP is about 120 MB, mostly due to the PDF.