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, 2 insertions, 2 deletions

diff --git a/include/pv.h b/include/pv.h
index 06455fe..1754829 100644
--- a/include/pv.h
+++ b/include/pv.h
@@ -27,11 +27,11 @@
 #define PV_MAX_CELLS (6 * 6)
 
 typedef struct pv_leaf_s {
-  float head[3];
+  float head[2];
   uint8_t cells[PV_MAX_CELLS];
-  uint8_t current_colour;
   uint8_t terminal;
   float terminal_value;
+  float parity_sign;
 } pv_leaf_t;
 
 // Capture a horizon leaf (winner == 0xFF) or an exact terminal leaf.