credit Mitchell Riley: we can drop 65 parameters for about the same strength; also train for longer

The idea is that the current model is M_theta(board, ply) = "score for player 0", but we can instead create a different model M'_theta(board) = "score for player 0 assuming that it's player 0's move" and for a certain pair of functions sigma, normalise, recover M_theta(board, ply) = sigma(ply) M'_theta(normalise(board, ply)).

This pair of functions is sigma(ply) = +/- 1 (parity), and normalise being a conditional interchange of pieces on the board depending on the parity of the ply. The only things stopping this from being completely straightforward is that nn_train.py evaluates M'_theta itself so we need to pass sigma(ply) to the training loop.

6x6 is still untrained.

1 files changed, 1 insertions, 1 deletions

diff --git a/include/weights_5.h b/include/weights_5.h
index 16a9465..8ad9d1b 100644
--- a/include/weights_5.h
+++ b/include/weights_5.h
@@ -18,7 +18,7 @@
 #ifndef WEIGHTS_5_H
 #define WEIGHTS_5_H
 
-#define FIVE_INP_NUM 3 + 5 * 5
+#define FIVE_INP_NUM (2 + 5 * 5)
 #define FIVE_DENSE_NUM 64
 
 // this does some kind of magic vectorisation