Something to note is that I needed to reconvert my dataset as the score was stm relative, Im pretty sure what I converted was the exact dataset with fixed scores, however it seems that a network trained with wdl 1 on the fixed data set loses a few elo: http://aytchell.eu.pythonanywhere.com/test/408/
Passed STC: Elo | 10.39 +- 7.05 (95%) SPRT | 8.0+0.08s Threads=1 Hash=8MB LLR | 2.90 (-2.25, 2.89) [0.00, 5.00] Games | N: 5016 W: 1426 L: 1276 D: 2314 Penta | [90, 545, 1105, 661, 107] http://aytchell.eu.pythonanywhere.com/test/413/
Something to note is that I needed to reconvert my dataset as the score was stm relative, Im pretty sure what I converted was the exact dataset with fixed scores, however it seems that a network trained with wdl 1 on the fixed data set loses a few elo: http://aytchell.eu.pythonanywhere.com/test/408/
bench 5435503