Call Trainer._setup() in constructor, to logically allow custom graph built outside trainer. (#318)

tensorpack/train/base.py

@@ -64,6 +64,8 @@ class Trainer(object):
         self.monitors = []
         self._epoch_num = None
 
+        self._setup()   # subclass will setup the graph
+
     @property
     def epoch_num(self):
         if self._epoch_num is not None:
@@ -76,9 +78,6 @@ class Trainer(object):
         """
         Use this method before :meth:`Trainer._setup` finishes,
         to register a callback to the trainer.
-
-        The hooks of the registered callback will be bind to the
-        `self.hooked_sess` session.
         """
         assert isinstance(cb, Callback), cb
         assert not isinstance(self._callbacks, Callbacks), \
@@ -120,8 +119,6 @@ class Trainer(object):
         """
         Setup the trainer and be ready for the main loop.
         """
-        self._setup()   # subclass will setup the graph
-
         self.register_callback(MaintainStepCounter())
         for cb in self.config.callbacks:
             self.register_callback(cb)
@@ -160,9 +157,15 @@ class Trainer(object):
         self.hooked_sess = tf.train.MonitoredSession(
             session_creator=ReuseSessionCreator(self.sess), hooks=hooks)
 
-    @abstractmethod
     def _setup(self):
-        """ setup Trainer-specific stuff for training"""
+        """
+        Build the entire graph for training.
+        Responsible for setup InputSource as well (including registering InputSource callbacks)
+
+        Since this method will get called in constructor only,
+        you can simply leave it empty and build your graph outside the trainer.
+        """
+        pass
 
     @property
     def global_step(self):