Pea - a UCI compliant chess engine written in Rust with a NNUE trained from scratch.
Author: WGCodings
Pea 7.0 what's new?
More improvements to NNUE.
Training parameters for this run :
Positions : 20M Net 0
10k nodes per position
Initial LR : 0.0005
Final LR : 0.0001 * 0.3f32.powi(5)
~2 epochs
start WDL : 0.0
end WDL : 0.15
Results vs previous generation
Net 0
Results of GEN7_NET0 vs GEN6_NET0 (8+0.08, 1t, 256MB, book.epd):
Elo: 102.28 +/- 22.59, nElo: 129.68 +/- 27.00
Games: 636, Wins: 309, Losses: 127, Draws: 200, Points: 409.0 (64.31 %)
Net 1
Results of GEN7_NET1 vs GEN6_NET1 (8+0.08, 1t, 256MB, book.epd):
Elo: 81.65 +/- 19.71, nElo: 106.71 +/- 24.80
Games: 754, Wins: 344, Losses: 170, Draws: 240, Points: 464.0 (61.54 %)
Net 1 Elo estimation ~2850 Elo
Results of GEN7_NET1 vs STASH24-2880 (8+0.08, 1t, 256MB, book.epd):
Elo: -29.94 +/- 13.26, nElo: -35.52 +/- 15.64
Games: 1896, Wins: 660, Losses: 823, Draws: 413, Points: 866.5 (45.70 %)
Net 1 included in binary.
Comments
Post a Comment