[tune] save error msg, cleanup after object checkpoints

python/ray/rllib/a3c/a3c.py

@@ -114,9 +114,9 @@ class A3CAgent(Agent):
 
         return result
 
-    def _save(self):
+    def _save(self, checkpoint_dir):
         checkpoint_path = os.path.join(
-            self.logdir, "checkpoint-{}".format(self.iteration))
+            checkpoint_dir, "checkpoint-{}".format(self.iteration))
         agent_state = ray.get(
             [a.save.remote() for a in self.remote_evaluators])
         extra_data = {

python/ray/rllib/agent.py

@@ -147,8 +147,8 @@ class _MockAgent(Agent):
             episode_reward_mean=10, episode_len_mean=10,
             timesteps_this_iter=10, info={})
 
-    def _save(self):
-        path = os.path.join(self.logdir, "mock_agent.pkl")
+    def _save(self, checkpoint_dir):
+        path = os.path.join(checkpoint_dir, "mock_agent.pkl")
         with open(path, 'wb') as f:
             pickle.dump(self.info, f)
         return path

python/ray/rllib/dqn/dqn.py

@@ -218,10 +218,10 @@ class DQNAgent(Agent):
         else:
             self.local_evaluator.sample(no_replay=True)
 
-    def _save(self):
+    def _save(self, checkpoint_dir):
         checkpoint_path = self.saver.save(
             self.local_evaluator.sess,
-            os.path.join(self.logdir, "checkpoint"),
+            os.path.join(checkpoint_dir, "checkpoint"),
             global_step=self.iteration)
         extra_data = [
             self.local_evaluator.save(),

python/ray/rllib/es/es.py

@@ -300,9 +300,9 @@ class ESAgent(Agent):
 
         return result
 
-    def _save(self):
+    def _save(self, checkpoint_dir):
         checkpoint_path = os.path.join(
-            self.logdir, "checkpoint-{}".format(self.iteration))
+            checkpoint_dir, "checkpoint-{}".format(self.iteration))
         weights = self.policy.get_weights()
         objects = [
             weights,

python/ray/rllib/ppo/ppo.py

@@ -244,10 +244,10 @@ class PPOAgent(Agent):
 
         return result
 
-    def _save(self):
+    def _save(self, checkpoint_dir):
         checkpoint_path = self.saver.save(
             self.local_evaluator.sess,
-            os.path.join(self.logdir, "checkpoint"),
+            os.path.join(checkpoint_dir, "checkpoint"),
             global_step=self.iteration)
         agent_state = ray.get(
             [a.save.remote() for a in self.remote_evaluators])