Expose SimpleTrainer.build_graph; Provide a default Trainer.run_step

examples/GAN/GAN.py

@@ -71,10 +71,10 @@ class GANTrainer(FeedfreeTrainerBase):
         opt = self.model.get_optimizer()
 
         # by default, run one d_min after one g_min
-        self.g_min = opt.minimize(self.model.g_loss, var_list=self.model.g_vars, name='g_op')
-        with tf.control_dependencies([self.g_min]):
-            self.d_min = opt.minimize(self.model.d_loss, var_list=self.model.d_vars, name='d_op')
-        self.train_op = self.d_min
+        g_min = opt.minimize(self.model.g_loss, var_list=self.model.g_vars, name='g_op')
+        with tf.control_dependencies([g_min]):
+            d_min = opt.minimize(self.model.d_loss, var_list=self.model.d_vars, name='d_op')
+        self.train_op = d_min
 
 
 class SeparateGANTrainer(FeedfreeTrainerBase):
@@ -137,12 +137,12 @@ class MultiGPUGANTrainer(MultiGPUTrainerBase, FeedfreeTrainerBase):
 
         opt = self.model.get_optimizer()
         # run one d_min after one g_min
-        self.g_min = opt.minimize(g_loss, var_list=self.model.g_vars,
+        g_min = opt.minimize(g_loss, var_list=self.model.g_vars,
                              colocate_gradients_with_ops=True, name='g_op')
-        with tf.control_dependencies([self.g_min]):
-            self.d_min = opt.minimize(d_loss, var_list=self.model.d_vars,
+        with tf.control_dependencies([g_min]):
+            d_min = opt.minimize(d_loss, var_list=self.model.d_vars,
                                  colocate_gradients_with_ops=True, name='d_op')
-        self.train_op = self.d_min
+        self.train_op = d_min
 
 
 class RandomZData(DataFlow):

tensorpack/callbacks/inference_runner.py

@@ -95,7 +95,8 @@ class InferenceRunnerBase(Callback):
 
         self._input_source.setup(self.trainer.model.get_inputs_desc())
         with tf.variable_scope(tf.get_variable_scope(), reuse=True):
-            self._tower_handle = self.trainer.predictor_factory.build(self._tower_name, device, self._input_source)
+            self._tower_handle = self.trainer.predictor_factory.build(
+                self._tower_name, device, self._input_source)
 
         self._hooks = [self._build_hook(inf) for inf in self.infs]
         cbs = self._input_source.get_callbacks()

tensorpack/graph_builder/predictor_factory.py

@@ -84,7 +84,7 @@ class PredictorFactory(object):
         Args:
             tower (int): need the kth tower (not the gpu id, but the id in TrainConfig.predict_tower)
         Returns:
-            an online predictor (which has to be used under the default session)
+            an online predictor (which has to be used under a default session)
         """
         tower_name = 'towerp{}'.format(tower)
         tower = self._towers[tower]

tensorpack/train/base.py

@@ -2,10 +2,8 @@
 # File: base.py
 # Author: Yuxin Wu <ppwwyyxxc@gmail.com>
 
-from abc import ABCMeta, abstractmethod
 import time
 import weakref
-import six
 from six.moves import range
 
 import tensorflow as tf
@@ -30,7 +28,6 @@ class StopTraining(BaseException):
     pass
 
 
-@six.add_metaclass(ABCMeta)
 class Trainer(object):
     """ Base class for a trainer.
 
@@ -102,10 +99,18 @@ class Trainer(object):
         self.setup()
         self.main_loop()
 
-    @abstractmethod
     def run_step(self):
-        """ Abstract method: run one iteration. Subclass should define what is "iteration".
         """
+        Defines what to do in one iteration, by default is:
+        ``self.hooked_sess.run(self.train_op)``.
+
+        The behavior can be changed by either defining what is ``train_op``,
+        or overriding this method.
+        """
+        assert hasattr(self, 'train_op'), \
+            "Please either set `Trainer.train_op` or provide an implementation " \
+            "of Trainer.run_step()!"
+        self.hooked_sess.run(self.train_op)
 
     def _setup_input_source(self, input_source):
         """

tensorpack/train/feedfree.py

@@ -29,10 +29,6 @@ class FeedfreeTrainerBase(Trainer):
         assert isinstance(self._input_source, FeedfreeInput), type(self._input_source)
         self._setup_input_source(self._input_source)
 
-    def run_step(self):
-        """ Simply run ``self.train_op``."""
-        self.hooked_sess.run(self.train_op)
-
 
 # deprecated
 class SingleCostFeedfreeTrainer(FeedfreeTrainerBase):

tensorpack/train/simple.py

@@ -15,7 +15,7 @@ __all__ = ['SimpleTrainer']
 class SimpleTrainer(Trainer):
     """ A naive single-tower single-cost demo trainer.
         Support both InputSource and DataFlow.
-        When DataFlow is given, the InputSource to be used will be ``FeedInput(df)``.
+        When DataFlow is given instead of InputSource, the InputSource to be used will be ``FeedInput(df)``.
     """
 
     def __init__(self, config):
@@ -35,13 +35,28 @@ class SimpleTrainer(Trainer):
                         "Consider QueueInput or other InputSource instead.")
         super(SimpleTrainer, self).__init__(config)
 
-    def run_step(self):
-        self.hooked_sess.run(self.train_op)
+    @staticmethod
+    def setup_graph(self, model, input):
+        """
+        Setup graph for simple trainer.
 
-    def _setup(self):
-        self._setup_input_source(self._input_source)
+        Args:
+            model (ModelDesc):
+            input (InputSource):
+
+        Returns:
+            tf.Operation: the training op
+            [Callback]: the callbacks to be added
+        """
+        input.setup(model.get_inputs_desc())
+        cbs = input.get_callbacks()
         with TowerContext('', is_training=True):
-            self.model.build_graph(self._input_source)
-            cost, grads = self.model.get_cost_and_grad()
-        opt = self.model.get_optimizer()
-        self.train_op = opt.apply_gradients(grads, name='min_op')
+            model.build_graph(input)
+            _, grads = model.get_cost_and_grad()
+        opt = model.get_optimizer()
+        train_op = opt.apply_gradients(grads, name='min_op')
+        return train_op, cbs
+
+    def _setup(self):
+        self.train_op, callbacks = SimpleTrainer.setup_graph(self.model, self._input_source)
+        self.config.callbacks.extend(callbacks)