TalkChess.com

Stockfish 101219 Vs Lc0 61441 Hardware 2950x, RTX 2080 ti TC = 0m + 300s Stockfish 101219 Contempt = 0 Threads = 32 SyzygyProbeDepth =1 Hash Size = 8 Gb Book: Perfect 2019, 8 moves Lc0 23.1 - 61441 Settings Customized Book: Perfect 2019, 8 moves Live Stream: https://www.youtube.com/watch?v=g2221A81...

Engines need to know that they're playing armageddon. So they need to be taught playing that mode. I agree, in the same way AlphaZero has been trained to play No Castling Chess, Lc0 must learn No Castling Chess or BNC with draw advantage, while alpha-beta programs must be also modified. A new openi...

It seems that AlphaZero has been trained to play "no-castling chess", see the interesting article

Kramnik And AlphaZero: How To Rethink Chess‎

at

https://www.chess.com/article/view/no-c ... -alphazero

I think that one important thing for testing chess engines is the different time control for each side.

As far as I know only cutechess-cli and aquarium have this feature.

Is it so difficult to program this?

Lc0BoostSFdevMyOp 2019 Stockfish_19032014_x64_bmi2 2200 - Lc0Boost32930 2200 49.0 - 51.0 +12/=74/-14 49.00% Stockfish_19032014_x64_bmi2 2200 - Lc0_41871 2200 44.5 - 55.5 +9/=71/-20 44.50% Stockfish_19032014_x64_bmi2 2200 - Lc0Boost7000_32930 2200 51.0 - 49.0 +11/=80/-9 51.00% Level: Blitz 5/3 Hardwa...

From your results it seems that the 1000 steps nets are better than the 7000 steps. From my results at longer time control the differences decrease.. Engine Score St 1: Stockfish_10_x64 32,5/66 ······················ 2: Lc0v41997 12,5/22 ===10===01==1=11=0===1 3: Lc0Boost7000_41800 11,0/22 ==0=0=0=0...

dkappe wrote: ↑

Sat Apr 20, 2019 8:55 pm

About even with 41800, but plays more aggressively, less "trolly."

https://github.com/dkappe/leela-chess-w ... t-Networks

Thanks for share, I will test the new net.

Here are my two running tournaments. It seems that LC0Boost1000 steps scores better than 7000 steps. Both with Silver Suite 50x2 games in each match Engine Score St 1: Stockfish_19032014_x64 104/222 ·········································································· 2: Lc0Boost32930 41,5/74 ...

Here are my two running tournaments. It seems that LC0Boost1000 steps scores better than 7000 steps. Both with Silver Suite 50x2 games in each match Engine Score St 1: Stockfish_19032014_x64 104/222 ·········································································· 2: Lc0Boost32930 41,5/74 =...

After some additional testing, the boost net at 7000 steps is about the same head 2 head and vs sf10 at 1+1. The style of play is a bit different, however. Would appreciate feedback. https://github.com/dkappe/leela-chess-weights/wiki/Boost-Networks Thank you very much! I am going to test the new Bo...