debugging wonky logging (#58)

jvmncs · Feb 21, 2019 · 6350d1b · 6350d1b
1 parent a25b9ab
commit 6350d1b
Show file tree

Hide file tree

Showing 3 changed files with 8 additions and 2 deletions.
diff --git a/safe_grid_agents/common/agents/policy_base.py b/safe_grid_agents/common/agents/policy_base.py
@@ -152,6 +152,9 @@ def gather_rollout(self, env, env_state, history, args) -> Rollout:
                 state = successor
                 history["t"] += 1
 
+            if r != 0:
+                history["episode"] += 1
+
             returns = self.get_discounted_returns(rewards)
             history = track_metrics(history, env)
             rollout.states.append(states)

diff --git a/safe_grid_agents/common/agents/policy_crmdp.py b/safe_grid_agents/common/agents/policy_crmdp.py
@@ -182,6 +182,9 @@ def gather_rollout(self, env, env_state, history, args) -> Rollout:
                 state = successor
                 history["t"] += 1
 
+            if r != 0:
+                history["episode"] += 1
+
             self.identify_corruption_in_trajectory(successors_r, rewards)
             rewards = self.get_modified_rewards_for_rollout(successors_r, rewards)
 

diff --git a/train.py b/train.py
@@ -56,7 +56,7 @@ def train(args, config=None, reporter=noop):
     agent, env, history, args = warmup_fn(agent, env, history, args)
 
     ######## Learn (and occasionally evaluate) ########
-    history["t"], eval_history["period"] = 0, 0
+    history["t"], history["episode"], eval_history["period"] = 0, 0, 0
 
     for episode in range(args.episodes):
         env_state = (
@@ -65,7 +65,7 @@ def train(args, config=None, reporter=noop):
             False,
             {"hidden_reward": 0.0, "observed_reward": 0.0},
         )
-        history["episode"] = episode
+        history["episode"] += 1
         env_state, history, eval_next = learn_fn(agent, env, env_state, history, args)
         info = env_state[3]
         reporter(