Python3.5対応

a3c.py

@@ -88,13 +88,13 @@ def log_uniform(lo, hi, rate):
 checkpoint = tf.train.get_checkpoint_state(CHECKPOINT_DIR)
 if checkpoint and checkpoint.model_checkpoint_path:
   saver.restore(sess, checkpoint.model_checkpoint_path)
-  print "checkpoint loaded:", checkpoint.model_checkpoint_path
+  print("checkpoint loaded:", checkpoint.model_checkpoint_path)
   tokens = checkpoint.model_checkpoint_path.split("-")
   # set global step
   global_t = int(tokens[1])
-  print ">>> global step set: ", global_t
+  print(">>> global step set: ", global_t)
 else:
-  print "Could not find old checkpoint"
+  print("Could not find old checkpoint")
 
 
 def train_function(parallel_index):

a3c_display.py

@@ -66,9 +66,9 @@ def choose_action(pi_values):
 checkpoint = tf.train.get_checkpoint_state(CHECKPOINT_DIR)
 if checkpoint and checkpoint.model_checkpoint_path:
   saver.restore(sess, checkpoint.model_checkpoint_path)
-  print "checkpoint loaded:", checkpoint.model_checkpoint_path
+  print("checkpoint loaded:", checkpoint.model_checkpoint_path)
 else:
-  print "Could not find old checkpoint"
+  print("Could not find old checkpoint")
 
 game_state = GameState(0, display=True, no_op_max=0)
 

a3c_training_thread.py

@@ -114,8 +114,8 @@ def process(self, sess, global_t, summary_writer, summary_op, score_input):
       values.append(value_)
 
       if (self.thread_index == 0) and (self.local_t % 100) == 0:
-        print "pi=", pi_
-        print " V=", value_
+        print("pi=", pi_)
+        print(" V=", value_)
 
       # process game
       self.game_state.process(action)
@@ -136,7 +136,7 @@ def process(self, sess, global_t, summary_writer, summary_op, score_input):
 
       if terminal:
         terminal_end = True
-        print "score=", self.episode_reward
+        print("score=", self.episode_reward)
 
         self._record_score(sess, summary_writer, summary_op, score_input,
                            self.episode_reward, global_t)
@@ -201,7 +201,7 @@ def process(self, sess, global_t, summary_writer, summary_op, score_input):
               feed_dict = { self.learning_rate_input: cur_learning_rate } )
 
     if (self.thread_index == 0) and (self.local_t % 100) == 0:
-      print "TIMESTEP", self.local_t
+      print("TIMESTEP", self.local_t)
 
     # return advanced local step size
     diff_local_t = self.local_t - start_local_t

a3c_visualize.py

@@ -55,9 +55,9 @@
 checkpoint = tf.train.get_checkpoint_state(CHECKPOINT_DIR)
 if checkpoint and checkpoint.model_checkpoint_path:
   saver.restore(sess, checkpoint.model_checkpoint_path)
-  print "checkpoint loaded:", checkpoint.model_checkpoint_path
+  print("checkpoint loaded:", checkpoint.model_checkpoint_path)
 else:
-  print "Could not find old checkpoint"
+  print("Could not find old checkpoint")
 
 W_conv1 = sess.run(global_network.W_conv1)
 

accum_trainer_test.py

@@ -58,7 +58,7 @@ def testBatchAccum(self):
       #cost = tf.square(mul)
       cost = tf.reduce_sum( tf.square(mul) )
 
-      print cost.get_shape()
+      print(cost.get_shape())
 
       trainer.prepare_minimize(cost, [var0])
 

game_state.py

@@ -10,13 +10,13 @@
 class GameState(object):
   def __init__(self, rand_seed, display=False, no_op_max=7):
     self.ale = ALEInterface()
-    self.ale.setInt('random_seed', rand_seed)
+    self.ale.setInt(b'random_seed', rand_seed)
     self._no_op_max = no_op_max
 
     if display:
       self._setup_display()
 
-    self.ale.loadROM(ROM)
+    self.ale.loadROM(ROM.encode('ascii'))
 
     # collect minimal action set
     self.real_actions = self.ale.getMinimalActionSet()
@@ -51,10 +51,10 @@ def _setup_display(self):
     if sys.platform == 'darwin':
       import pygame
       pygame.init()
-      self.ale.setBool('sound', False)
+      self.ale.setBool(b'sound', False)
     elif sys.platform.startswith('linux'):
-      self.ale.setBool('sound', True)
-    self.ale.setBool('display_screen', True)
+      self.ale.setBool(b'sound', True)
+    self.ale.setBool(b'display_screen', True)
 
   def reset(self):
     self.ale.reset_game()