🧠 NN: Aurora-855bc8d7cb

[TheBlackPlague]

🎯 Summary

This PR switches the default neural network architecture from Starshard to Aurora. Aurora is a recently introduced architecture that is bigger than the Starshard architecture (Accumulator size change from 256 to 384), and, thus, does slow down the neural network updating and evaluation a bit. However, with the increased accuracy, it triumphs the current Starshard network.

🧬 Architecture and Hyper-Parameter

       IN          ACCUMULATOR                               HIDDEN                                    OUT
 ______________      _______      ______________________________________________________________      _____
| WHITE: (768) | -> | (384) | -> | ClippedReLU(0, 1) -> (384) \                                 |    |     |
|              |    |       |    |                             \                                |    |     |
|              |    |       |    |                              CONCATENATE(ColorToMove): (768) | -> | (1) |
|              |    |       |    |                             /                                |    |     |
| BLACK: (768) | -> | (384) | -> | ClippedReLU(0, 1) -> (384) /                                 |    |     |
 --------------      -------      --------------------------------------------------------------      -----

Codename: Aurora
ID: 855bc8d7cb

Data: 1.7B FEN --- DEPTH: 9 (1.2B) + NODES: 5K (500M)
Batch Size: 500K
Epochs: 84
LR: 7e-3
LR Drop Step: 1
LR Drop Last: 120
LR Gamma: 0.96
WDL: 0
Scale: 300 (Inference: 400)

👏 Acknowledgements

NA

📈 ELO

STC:

ELO   | 5.82 +- 4.50 (95%)
SPRT  | 10.0+0.10s Threads=1 Hash=16MB
LLR   | 3.00 (-2.94, 2.94) [0.00, 5.00]
GAMES | N: 14568 W: 4765 L: 4521 D: 5282

LTC:

ELO   | 5.19 +- 4.03 (95%)
SPRT  | 60.0+0.60s Threads=1 Hash=256MB
LLR   | 3.04 (-2.94, 2.94) [0.00, 5.00]
GAMES | N: 16600 W: 4973 L: 4725 D: 6902