Merge pull request #27 from wniec/update-normalize-order-of-updates

fix: update after normalization

Author: wniec

dynamicalgorithmselection/agents/agent_utils.py

@@ -135,12 +135,12 @@ def __init__(self, max_steps, clip_reward=10.0):
 
     def normalize(self, reward, step_idx, update=True):
         idx = min(step_idx, self.max_steps - 1)
-        if update:
-            self.stats[idx].update(np.array([reward]))
 
         mean = self.stats[idx].mean
         std = np.sqrt(self.stats[idx].var) + 1e-8
-
         normalized_reward = (reward - mean) / std
 
+        if update:
+            self.stats[idx].update(np.array([reward]))
+
         return np.clip(normalized_reward, -self.clip, self.clip)