TalkChess.com

My best one is 2141 from 22.07, one before yours. Not enough games: 20+0.1 # PLAYER : RATING ERROR POINTS PLAYED (%) 1 stnnuesergio20200723-0511 : 2857.0 2.9 1706.5 3393 50.3% 2 stnnuesergio20200722-2141 : 2855.0 2.9 1686.5 3393 49.7%

Someone tested experimental nets? I don't see many comments about them.

2344 For me the best net was 20200723-0511. Comparing it against 20200724-2344 after 1500 games 20+0.1 probably does not change this: # PLAYER : RATING ERROR POINTS PLAYED (%) 1 stnnuesergio20200723-0511 : 2859.0 4.5 792.0 1558 50.8% 2 stnnuesergio20200724-2344 : 2853.0 4.5 766.0 1558 49.2%

Why hasn't anybody asked noobpwnftw of ChessDB fame for his data? I think he has more than 10 billion positions analyzed with some 22 Depth or something, maybe training from that would be good. Why hasn't anybody trained a NNUE based on Leela's eval? There's people without GPUs that haven't been ab...

Stockfish Dev 1706 vs sergio 20200722-1517 200 + 2: # PLAYER : RATING ERROR POINTS PLAYED (%) 1 stnnuesergio20200722-1517 : 2886.5 10.8 160.5 274 58.6% 2 st200617 : 2825.5 10.8 113.5 274 41.4% Games: https://1drv.ms/u/s!Aj8wlSmr_Hq3irYFs2REWTBwPoJkdg?e=BTECXZ

A fast test between two of the latest sergio's nets, 15+0.1 # PLAYER : RATING ERROR POINTS PLAYED (%) 1 20200722-1517 : 2866.5 7.2 340.5 643 53.0% 2 20200722-0359 : 2845.5 7.2 302.5 643 47.0% Same best sergio's net against GK best (that was already better than Stockfish dev), 20 + 0.1: # PLAYER : RA...

Hello. I have done a test between SFNNUE GK 27-6 against Sergio's network. More games needed to differentiate between them. 200+1 # PLAYER : RATING ERROR POINTS PLAYED (%) 1 stnnuesergio2007 : 2857.8 4.7 303.5 601 50.5% 2 stnnuegk2706 : 2854.2 4.7 297.5 601 49.5% Games: https://1drv.ms/u/s!Aj8wlSmr_...

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...