Stingray (5.6.0.0)
What's Changed
- Templated Make Move for Normal, Classical, and NN Updates. by @TheBlackPlague in #83
- Update History Heuristics on Beta Cutoffs. by @TheBlackPlague in #84
- Increase neural network training batch size & reliance on game result. by @TheBlackPlague in #85
NNUE:
IN ACCUMULATOR HIDDEN OUT
______________ _______ ______________________________________________________________ _____
| WHITE: (768) | -> | (256) | -> | ClippedReLU(0, 1) -> (256) \ | | |
| | | | | \ | | |
| | | | | CONCATENATE(ColorToMove): (512) | -> | (1) |
| | | | | / | | |
| BLACK: (768) | -> | (256) | -> | ClippedReLU(0, 1) -> (256) / | | |
-------------- ------- -------------------------------------------------------------- -----
Data: 1.7B FEN --- DEPTH: 9 (1.2B) + NODES: 5K (500M)
Epochs: 30
WDL: 0.4
Batch Size: 500K
Full Changelog: 5.5.0.0...5.6.0.0
ELO Difference
Using UHO_XXL_+0.90_+1.19.epd:
TC: 10s + 0.1s (STC)
ELO | 107.31 +- 5.72 (95%)
CONF | 10.0+0.10s Threads=1 Hash=16MB
GAMES | N: 10000 W: 4845 L: 1851 D: 3304
TC: 60s + 0.6s (LTC)
ELO | 110.14 +- 7.68 (95%)
CONF | 60.0+0.60s Threads=1 Hash=256MB
GAMES | N: 5000 W: 2294 L: 760 D: 1946
Contributors
TheBlackPlague