fix some atari settings

examples/Atari2600/atari.py

@@ -62,8 +62,7 @@ class AtariPlayer(RLEnvironment):
         with _ALE_LOCK:
             self.ale = ALEInterface()
             self.rng = get_rng(self)
-
-            self.ale.setInt(b"random_seed", self.rng.randint(0, 10000))
+            self.ale.setInt(b"random_seed", self.rng.randint(0, 30000))
             self.ale.setBool(b"showinfo", False)
 
             self.ale.setInt(b"frame_skip", 1)
@@ -132,7 +131,8 @@ class AtariPlayer(RLEnvironment):
 
     def restart_episode(self):
         self.current_episode_score.reset()
-        self.ale.reset_game()
+        with _ALE_LOCK:
+            self.ale.reset_game()
 
         # random null-ops start
         n = self.rng.randint(self.nullop_start)
@@ -160,11 +160,12 @@ class AtariPlayer(RLEnvironment):
 
         self.current_episode_score.feed(r)
         isOver = self.ale.game_over()
+        if self.live_lost_as_eoe:
+            isOver = isOver or newlives < oldlives
         if isOver:
             self.finish_episode()
+        if self.ale.game_over():
             self.restart_episode()
-        if self.live_lost_as_eoe:
-            isOver = isOver or newlives < oldlives
         return (r, isOver)
 
 if __name__ == '__main__':

examples/Atari2600/common.py

@@ -48,10 +48,11 @@ def eval_with_funcs(predict_funcs, nr_eval):
             return self._func(*args, **kwargs)
 
         def run(self):
-            player = get_player()
+            player = get_player(train=False)
             while not self.stopped():
                 try:
                     score = play_one_episode(player, self.func)
+                    #print "Score, ", score
                 except RuntimeError:
                     return
                 self.queue_put_stoppable(self.q, score)

tensorpack/RL/simulator.py

@@ -14,6 +14,7 @@ import numpy as np
 import six
 from six.moves import queue
 
+from ..models._common import disable_layer_logging
 from ..callbacks import Callback
 from ..tfutils.varmanip import SessionUpdate
 from ..predict import OfflinePredictor
@@ -221,6 +222,7 @@ class SimulatorProcessSharedWeight(SimulatorProcessDF):
         self.pred_config = pred_config
 
     def _prepare(self):
+        disable_layer_logging()
         self.predictor = OfflinePredictor(self.pred_config)
         with self.predictor.graph.as_default():
             vars_to_update = self._params_to_update()
@@ -244,6 +246,7 @@ class SimulatorProcessSharedWeight(SimulatorProcessDF):
     def _trigger_evt(self):
         with self.weight_lock:
             self.sess_updater.update(self.shared_dic['params'])
+            logger.info("Updated.")
 
     def _params_to_update(self):
         # can be overwritten to update more params
@@ -262,7 +265,12 @@ class WeightSync(Callback):
         # can be overwritten to update more params
         return tf.trainable_variables()
 
+    def _before_train(self):
+        self._sync()
     def _trigger_epoch(self):
+        self._sync()
+
+    def _sync(self):
         logger.info("Updating weights ...")
         dic = {v.name: v.eval() for v in self.vars}
         self.shared_dic['params'] = dic

tensorpack/train/trainer.py

@@ -117,6 +117,9 @@ class EnqueueThread(threading.Thread):
             try:
                 while True:
                     for dp in self.dataflow.get_data():
+                        #import IPython;
+                        #IPython.embed(config=IPython.terminal.ipapp.load_default_config())
+
                         if self.coord.should_stop():
                             return
                         feed = dict(zip(self.input_vars, dp))