#brainlearn — Public Fediverse posts

To remove entries with zero depth from a #BrainLearn experience file you may do: "jja dump experience.exp | perl -F, -lane 'print if $F[1] > 0' | jja restore experience-nonull.exp" #jja #chess

To remove entries with zero depth from a #BrainLearn experience file you may do: "jja dump experience.exp | perl -F, -lane 'print if $F[1] > 0' | jja restore experience-nonull.exp" #jja #chess

The last #jja bug concerning #BrainLearn experience files was incorrect deserializing of promotion moves. I have found, corrected the bug and tested it with all types of promotions. It works beautifully now! https://git.sr.ht/~alip/jja/commit/02e9aa49c8ff79434efbb57bb18f1097f1872438 Time to release a new #jja version!

The last #jja bug concerning #BrainLearn experience files was incorrect deserializing of promotion moves. I have found, corrected the bug and tested it with all types of promotions. It works beautifully now! https://git.sr.ht/~alip/jja/commit/02e9aa49c8ff79434efbb57bb18f1097f1872438 Time to release a new #jja version!

Too bad #pgn-extract has a flag --underpromotions to find games with underpromotions however no flag to find games with _any_ type of promotion. #grep to the rescue! "grep -Ei '=[QRBN]' caissa-ai.pgn -B18 > promotions.pgn" finds a game with any type of promotion and prints the whole #PGN. The good thing is this way I found a promotion entry in a #BrainLearn experience file and I found a bug in #jja's promotion handling!

Too bad #pgn-extract has a flag --underpromotions to find games with underpromotions however no flag to find games with _any_ type of promotion. #grep to the rescue! "grep -Ei '=[QRBN]' caissa-ai.pgn -B18 > promotions.pgn" finds a game with any type of promotion and prints the whole #PGN. The good thing is this way I found a promotion entry in a #BrainLearn experience file and I found a bug in #jja's promotion handling!

A reasonable theoretical opening line with an en-passant capture goes like this: 1.d4 f5 2.c4 Nf6 3.Nc3 g6 4.g3 Bg7 5.Bg2 d6 6.Nf3 O-O 7.O-O c6 8.d5 e5 9.dxe6 e.p. Thanks to kosmik on #chess channel at #Libera for remembering that! #jja #git can correctly deserialize this position and the en-passant capture from a #BrainLearn experience file.

A reasonable theoretical opening line with an en-passant capture goes like this: 1.d4 f5 2.c4 Nf6 3.Nc3 g6 4.g3 Bg7 5.Bg2 d6 6.Nf3 O-O 7.O-O c6 8.d5 e5 9.dxe6 e.p. Thanks to kosmik on #chess channel at #Libera for remembering that! #jja #git can correctly deserialize this position and the en-passant capture from a #BrainLearn experience file.

#jja's deserialization of en-passant and castling moves are tested to work for #BrainLearn experience files. Now I need to find a reasonable opening line that has a promotion so as to test if promotion is deserialized correctly. Then we can release a new version :-)

#jja's deserialization of en-passant and castling moves are tested to work for #BrainLearn experience files. Now I need to find a reasonable opening line that has a promotion so as to test if promotion is deserialized correctly. Then we can release a new version :-)

I have also fixed #jja's deserializing for castling moves of #BrainLearn experience file entries. Now I have to test en-passant and promotion before releasing a new version, however those two are relatively hard to find in opening books than castling. So I have also added #BrainLearn to #pgn export support to #jja edit subcommand.

I have also fixed #jja's deserializing for castling moves of #BrainLearn experience file entries. Now I have to test en-passant and promotion before releasing a new version, however those two are relatively hard to find in opening books than castling. So I have also added #BrainLearn to #pgn export support to #jja edit subcommand.

#jja #git now has proper support for #Stockfish compatible #Zobrist hashes and querying #BrainLearn files works as expected. I'll release this as 0.6.1 after some testing.

#jja #git now has proper support for #Stockfish compatible #Zobrist hashes and querying #BrainLearn files works as expected. I'll release this as 0.6.1 after some testing.

Turns out #BrainLearn experience files use #Zobrist hashes but with different seends than #Polyglot. I'm working on implementing #Stockfish compatible #Zobrist hashes in #jja, it is almost done.

Turns out #BrainLearn experience files use #Zobrist hashes but with different seends than #Polyglot. I'm working on implementing #Stockfish compatible #Zobrist hashes in #jja, it is almost done.

du -hs * | sort -h | tail -n 10 | awk '{print $2}' | while read f; do df -h .; : > $f; done # truncate ten biggest files in a directory (sorry i have gone a bit lax with the wildcard, adjust as necessary). I have just used this to free 2T space which I am going to download later when I put https://caissa.ai bot back online. But for now we have our #Zobrist hash experiment and #BrainLearn book exploration

du -hs * | sort -h | tail -n 10 | awk '{print $2}' | while read f; do df -h .; : > $f; done # truncate ten biggest files in a directory (sorry i have gone a bit lax with the wildcard, adjust as necessary). I have just used this to free 2T space which I am going to download later when I put https://caissa.ai bot back online. But for now we have our #Zobrist hash experiment and #BrainLearn book exploration

So as my first attempt I have generated all unique #Zobrist hash keys from a database of #caissa-ai's games. https://caissa.ai is my lichess bot and uses a #BrainLearn experience file which grew up to roughly 5 million entries over time. The funny bit is none of the positions in caissa-ai's games match any of the #Zobrist keys in the #BrainLearn experience file. How is this even possible?? Next step: compile a huge database of #Zobrist keys in an efficient manner

So as my first attempt I have generated all unique #Zobrist hash keys from a database of #caissa-ai's games. https://caissa.ai is my lichess bot and uses a #BrainLearn experience file which grew up to roughly 5 million entries over time. The funny bit is none of the positions in caissa-ai's games match any of the #Zobrist keys in the #BrainLearn experience file. How is this even possible?? Next step: compile a huge database of #Zobrist keys in an efficient manner

so the newest version of #jja, which is 0.6.0, has read access for #BrainLearn experience files. Yet I have a riddle. I could not find a way to test the query interface because given a #BrainLearn experience entry, which is #Zobrist hash, move, depth, score, and performance, i don't have enough info to constitute a legit chess position. #Zobrist hash works one way as all hash functions do...

so the newest version of #jja, which is 0.6.0, has read access for #BrainLearn experience files. Yet I have a riddle. I could not find a way to test the query interface because given a #BrainLearn experience entry, which is #Zobrist hash, move, depth, score, and performance, i don't have enough info to constitute a legit chess position. #Zobrist hash works one way as all hash functions do...