This helps me avoid NaN when I have small rewards

2026-09-09 11:13:47 +08:00 · 2017-11-01 18:58:03 +08:00
parent 9ef309696f
commit 662c0834f6
1 changed files with 1 additions and 1 deletions
@@ -61,7 +61,7 @@ class DDPGAgent:
            done = (done or (config.max_episode_length and steps >= config.max_episode_length))
            next_state = self.state_normalizer(next_state)
            total_reward += reward
-            # reward = self.reward_normalizer(reward)
+            reward = self.reward_normalizer(reward)

            if not deterministic:
                self.replay.feed([state, action, reward, next_state, int(done)])