fixed screen dims bug in environment wrapper; python3 compatibility

Author: marctuscher

dqn/agent.py

@@ -1,4 +1,5 @@
 from __future__ import print_function
+from functools import reduce
 import os
 import time
 import random
@@ -224,7 +225,7 @@ def build_dqn(self):
 
       q_summary = []
       avg_q = tf.reduce_mean(self.q, 0)
-      for idx in xrange(self.env.action_size):
+      for idx in range(self.env.action_size):
         q_summary.append(tf.summary.histogram('q/%s' % idx, avg_q[idx]))
       self.q_summary = tf.summary.merge(q_summary, 'q_summary')
 
@@ -325,7 +326,7 @@ def build_dqn(self):
 
     tf.initialize_all_variables().run()
 
-    self._saver = tf.train.Saver(self.w.values() + [self.step_op], max_to_keep=30)
+    self._saver = tf.train.Saver(list(self.w.values()) + [self.step_op], max_to_keep=30)
 
     self.load_model()
     self.update_target_q_network()
@@ -373,7 +374,7 @@ def play(self, n_step=10000, n_episode=100, test_ep=None, render=False):
       self.env.env.monitor.start(gym_dir)
 
     best_reward, best_idx = 0, 0
-    for idx in xrange(n_episode):
+    for idx in range(n_episode):
       screen, reward, action, terminal = self.env.new_random_game()
       current_reward = 0
 

dqn/environment.py

@@ -11,7 +11,7 @@ def __init__(self, config):
         config.screen_width, config.screen_height, config.action_repeat, config.random_start
 
     self.display = config.display
-    self.dims = (screen_width, screen_height)
+    self.dims = (screen_height, screen_width)
 
     self._screen = None
     self.reward = 0
@@ -26,7 +26,7 @@ def new_game(self, from_random_game=False):
 
   def new_random_game(self):
     self.new_game(True)
-    for _ in xrange(random.randint(0, self.random_start - 1)):
+    for _ in range(random.randint(0, self.random_start - 1)):
       self._step(0)
     self.render()
     return self.screen, 0, 0, self.terminal
@@ -70,7 +70,7 @@ def act(self, action, is_training=True):
     cumulated_reward = 0
     start_lives = self.lives
 
-    for _ in xrange(self.action_repeat):
+    for _ in range(self.action_repeat):
       self._step(action)
       cumulated_reward = cumulated_reward + self.reward