Spaces:

henribonamy
/

chess-puzzles-training

Paused

henribonamy commited on Mar 25

Commit

1bc17f7

verified ·

1 Parent(s): f9ea845

Upload src/rl_training/rewards.py with huggingface_hub

Files changed (1) hide show

src/rl_training/rewards.py CHANGED Viewed

@@ -230,7 +230,7 @@ def compute_binary_rewards(
         elif non_obvious:
             reward = 0.8
         elif eval_reversal:
-            reward = -0.3
         else:
             if w_deep > 0.65:
                 proximity = max(0.0, 0.65 - w2_deep) / 0.65

         elif non_obvious:
             reward = 0.8
         elif eval_reversal:
+            reward = 0.0
         else:
             if w_deep > 0.65:
                 proximity = max(0.0, 0.65 - w2_deep) / 0.65