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/src/match.c b/src/match.c
index 6ae9fa7..459fab5 100644
--- a/src/match.c
+++ b/src/match.c
@@ -40,7 +40,7 @@ static inline float five_cand(tak_state_p state) {
   nn_prepare_input(cur_board, state, cand_five_embedding);
   return eval_net(cur_board, FIVE_INP_NUM, &cand_five_dense_weights[0][0],
                   cand_five_dense_biases, cand_five_logit_weights, cand_five_logit_bias,
-                  FIVE_DENSE_NUM);
+                  FIVE_DENSE_NUM, state->ply & 1 ? 1.0f : -1.0f);
 }
 
 static inline float six_cand(tak_state_p state) {
@@ -49,7 +49,7 @@ static inline float six_cand(tak_state_p state) {
   nn_prepare_input(cur_board, state, cand_six_embedding);
   return eval_net(cur_board, SIX_INP_NUM, &cand_six_dense_weights[0][0],
                   cand_six_dense_biases, cand_six_logit_weights, cand_six_logit_bias,
-                  SIX_DENSE_NUM);
+                  SIX_DENSE_NUM, state->ply & 1 ? 1.0f : -1.0f);
 }