updated model

my_model.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2f023b0292ff0d225d43e005826d45ce4e0f24ef202bbc1ba08e6f1960ffcc8
-size 2400942
+oid sha256:0f6f53bbf5b346674f37ce2eb18bbe4ce2eafe9b934ac9c44df2f289e1efa2c1
+size 6980981

reward_wrapper.py

@@ -19,7 +19,7 @@ class CustomRewardWrapper(gym.Wrapper):
         heights = obs["heights"]
 
         # # Default reward
-        reward = 2
+        reward = 1
         # # reward = ((self.height - max(heights)) / self.height)
         # # reward += np.sum(board)
         # reward = (self.height - max(heights)) / self.height
@@ -37,6 +37,9 @@ class CustomRewardWrapper(gym.Wrapper):
         #
         # # Penalty for big differences between columns
         reward -= self.get_bumpiness_and_height(board)[0] / self.height
+
+        # Penalty for high columns
+        reward += (self.height / 2 ) - np.max(heights) * 0.5
         #
         # # Penalty for holes
         # # holes_val = 0