search

pkrisz99 / Renegade

A strong chess engine using a neural network to evaluate positions
https://lichess.org/@/RenegadeEngine
MIT License
24 stars 2 forks source link

Net 20: include 200 million DFRC positions #29

Closed pkrisz99 closed 6 months ago

pkrisz99 commented 6 months ago

First iteration of the net in DFRC:

Elo   | 117.39 +- 33.79 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | 2.91 (-2.25, 2.89) [0.00, 10.00]
Games | N: 258 W: 117 L: 33 D: 108
Penta | [0, 12, 46, 46, 25]
https://renegadedev.eu.pythonanywhere.com/test/89/

Second iteration in DFRC, with DFRC data being fully regenerated:

Elo   | 27.01 +- 10.77 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | 2.91 (-2.25, 2.89) [0.00, 5.00]
Games | N: 1766 W: 458 L: 321 D: 987
Penta | [28, 155, 398, 256, 46]
https://renegadedev.eu.pythonanywhere.com/test/92/

Passed non-regression for standard chess:

Elo   | 1.06 +- 3.74 (95%)
SPRT  | 8.0+0.08s Threads=1 Hash=16MB
LLR   | 2.97 (-2.94, 2.94) [-5.00, 0.00]
Games | N: 16084 W: 3932 L: 3883 D: 8269
Penta | [95, 1944, 3933, 1957, 113]
https://renegadedev.eu.pythonanywhere.com/test/91/

Total dataset size now: 2.249.670.917 Version format changed to avoid potential confusion

Renegade dev 1.0.25 Bench: 3598564