TalkChess.com

Laskos wrote: ↑

Mon Jul 20, 2020 1:36 pm

I don't have much endgame suites, but positionally the result seems to confirm a lesser accent on openings, unlike Lc0 zero nets.

Yes. Stockfish evaluation of openings is not specially good. So another nice try can be to mix somehow Lco evaluations to train the nets.

Maybe a depth 1 match between LC0 and NNUE will be useful, to get an idea of how the networks perform against each other, and of what importance the whole search is, or alike. -- Srdja No, the search is still SF. To depth=1 I compared SF NNUE to SF_dev, and SF NNUE is significantly stronger to dept...

Ok, thanks, first assessment is a bit disappointing. There's no winning game by either side game that's over before move 80, median game length is 111 ply for SF_NNUE and 118 ply for SF_dev, but actually it's longer than that because you're starting from FENs, and cutechess is counting from the FEN...

Congratulations! Is really very strong, stronger than regular Stockfish. I'm doing a little test at 5 minutes + 2 seconds and is wining also, like the other tests published here.

Thanks for StockFinn!! It already helped me to solve a very deep position

I've updated the net again: https://drive.google.com/file/d/1AjGfxPxDN-7NoflZu8wPRaeAgSNlhnXC/view?usp=sharing Tested +25 +/-8 elo vs stnnuegk2 after 2000 games. TC:10s+0.1s. 1 thread. Slowmover 50 for both Thanks. I have done two tests: 40 seconds + 0.3 against previous version: # PLAYER : RATING ...

Trying to understand this but not much luck so far. I run sf-nnue run (halfkp-256x2-32-32) x64-bmi2 from the command line and then used go depth 20 The engine just stopped after some time without any output, is this expected. Suggestion - allow nn.bin to be in same directory as the exe, this seems ...

Problably new best net, also from Gekkehenker: 40+0.3 # PLAYER : RATING ERROR POINTS PLAYED (%) 1 stnnuegk2 : 2860.6 3.9 930.5 1814 51.3% 2 stnnuegk1 : 2851.4 3.9 883.5 1814 48.7% Here the net: https://cdn.discordapp.com/attachments/718853716266188890/725417723609481296/nn_best_24_june.7z Thanks to ...

New test at longer time control 200 + 0.5

Gekkehenker comments: TB6 + draws included in the trainging data. Almost 1 billion depth 8 positions with TB6. gensfen depth 8 loop 1000000000 random_max_ply 14 random_move _count 7 write_min_ply 14 output_file_name 1bdepth8.bin Command to learn: learn targetdir traindir loop 100 batchsize 1000000 e...