fixes for cleanliness

README.md

@@ -1,5 +1,7 @@
 # pytorch-a3c
 
+xvfb-run -s "-screen 0 1400x900x24" python main.py
+
 This is a PyTorch implementation of Asynchronous Advantage Actor Critic (A3C) from ["Asynchronous Methods for Deep Reinforcement Learning"](https://arxiv.org/pdf/1602.01783v1.pdf).
 
 This implementation is inspired by [Universe Starter Agent](https://github.com/openai/universe-starter-agent).

main.py

@@ -45,6 +45,10 @@
     env = create_car_racing_env()
     model = ActorCritic(env.observation_space.shape[0], env.action_space)
 
+    itr = 0
     while True:
+        if itr % 10 == 0:
+            print ("Itr: ", itr)
         train(args, model, env)
         test(args, model, env)
+        itr += 1

test.py

@@ -35,7 +35,7 @@ def test(args, model, env):
     actions = deque(maxlen=100)
     episode_length = 0
     while True:
-        env.render()
+        #env.render()
         episode_length += 1
         # Sync with the shared model
         if done:

train.py

@@ -33,7 +33,7 @@ def train(args, model, env, optimizer=None):
     episode_length = 0
     u = 0
     while u < args.num_updates:
-        print ("update: ", u)
+        #print ("update: ", u)
         episode_length += 1
         # Sync with the shared model
         # model.load_state_dict(shared_model.state_dict())