fix bug: omit last action during evaluation

Felixvillas · Aug 25, 2022 · 2617bda · 2617bda
1 parent 89a4a6d
commit 2617bda
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/model.py b/model.py
@@ -297,10 +297,11 @@ def evaluate(self, env, episode_num=32):
         for _ in range(episode_num):
             eval_ep_rewards = []
             done = False
+            action = None
             env.reset()
             self.Q.init_eval_rnn_hidden()
             while not done:
-                last_obs = env.get_obs()
+                last_obs = env.get_obs(action)
                 avail_actions = env.get_avail_actions()
                 recent_observations = np.concatenate([np.expand_dims(ob, axis=0) for ob in last_obs], axis=0)
                 random_selection = np.zeros(self.num_agents).astype(np.bool_)