make 'TrainConfig' a simple key-value holder -- let it use train_with_defaults...

make 'TrainConfig' a simple key-value holder -- let it use train_with_defaults to handle the defaults

examples/DeepQNetwork/DQNModel.py

 #!/usr/bin/env python
 # -*- coding: utf-8 -*-
 # File: DQNModel.py
-# Author: Yuxin Wu <ppwwyyxxc@gmail.com>
+
 
 import abc
 import tensorflow as tf

examples/ImageNetModels/README.md

@@ -28,7 +28,7 @@ Evaluate the [pretrained model](http://models.tensorpack.com/ShuffleNet/):
 
 This Inception-BN script reaches 27% single-crop error after 300k steps with 6 GPUs.
 
-This VGG16 script reaches 28.8% single-crop error after 100 epochs (30h with 8 P100s). It gets 1% better if BN is enabled.
+This VGG16 script reaches 29~30% single-crop error after 100 epochs (30h with 8 P100s), and 28% if BN is enabled.
 
 ### ResNet, DoReFa-Net
 

examples/ImageNetModels/vgg16.py

@@ -124,7 +124,7 @@ if __name__ == '__main__':
     if args.gpu:
         os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
 
-    logger.set_logger_dir(os.path.join('train_log', 'vgg16'))
+    logger.set_logger_dir(os.path.join('train_log', 'vgg16-norm={}'.format(args.norm)))
 
     config = get_config()
     if args.load:

tensorpack/predict/config.py

@@ -18,31 +18,37 @@ class PredictConfig(object):
                  model=None,
                  inputs_desc=None,
                  tower_func=None,
-                 session_creator=None,
-                 session_init=None,
+
                  input_names=None,
                  output_names=None,
+
+                 session_creator=None,
+                 session_init=None,
                  return_input=False,
                  create_graph=True,
                  ):
         """
+        You need to set either `model`, or `inputs_desc` plus `tower_func`.
+        They are needed to construct the graph.
+        You'll also have to set `output_names` as it does not have a default.
+
         Args:
             model (ModelDescBase): to be used to obtain inputs_desc and tower_func.
             inputs_desc ([InputDesc]):
-            tower_func: a callable which takes input tensors
+            tower_func: a callable which takes input tensors and construct a tower.
+
+            input_names (list): a list of input tensor names. Defaults to match inputs_desc.
+            output_names (list): a list of names of the output tensors to predict, the
+                tensors can be any computable tensor in the graph.
 
             session_creator (tf.train.SessionCreator): how to create the
                 session. Defaults to :class:`tf.train.ChiefSessionCreator()`.
             session_init (SessionInit): how to initialize variables of the session.
                 Defaults to do nothing.
-            input_names (list): a list of input tensor names. Defaults to match inputs_desc.
-            output_names (list): a list of names of the output tensors to predict, the
-                tensors can be any computable tensor in the graph.
+
             return_input (bool): same as in :attr:`PredictorBase.return_input`.
             create_graph (bool): create a new graph, or use the default graph
                 when predictor is first initialized.
-
-        You need to set either `model`, or `inputs_desc` plus `tower_func`.
         """
         def assert_type(v, tp):
             assert isinstance(v, tp), v.__class__

tensorpack/train/base.py

@@ -7,6 +7,7 @@ import weakref
 import time
 from six.moves import range
 import six
+import copy
 
 from ..callbacks import (
     Callback, Callbacks, Monitors, TrainingMonitor)
@@ -187,6 +188,8 @@ class Trainer(object):
             callbacks ([Callback]):
             monitors ([TrainingMonitor]):
         """
+        assert isinstance(callbacks, list), callbacks
+        assert isinstance(monitors, list), monitors
         describe_trainable_vars()   # TODO weird
 
         self.register_callback(MaintainStepCounter())
@@ -284,7 +287,7 @@ class Trainer(object):
               session_creator, session_init,
               steps_per_epoch, starting_epoch=1, max_epoch=9999999):
         """
-        Implemented by:
+        Implemented by three lines:
 
         .. code-block:: python
 
@@ -299,18 +302,24 @@ class Trainer(object):
         self.main_loop(steps_per_epoch, starting_epoch, max_epoch)
 
     def train_with_defaults(
-            self, callbacks=None, monitors=None,
+            self, _sentinel=None,
+            callbacks=None, monitors=None,
             session_creator=None, session_init=None,
-            steps_per_epoch=None, starting_epoch=1, max_epoch=9999999):
+            steps_per_epoch=None, starting_epoch=1, max_epoch=9999999,
+            extra_callbacks=None):
         """
-        Same as :meth:`train()`, but will:
+        Same as :meth:`train()`, except:
 
-        1. Append :meth:`DEFAULT_CALLBACKS()` to callbacks.
-        2. Append :meth:`DEFAULT_MONITORS()` to monitors.
+        1. Add `extra_callbacks` to callbacks. The default value for
+           `extra_callbacks` is :meth:`DEFAULT_CALLBACKS()`.
+        2. Default value for `monitors` is :meth:`DEFAULT_MONITORS()`.
         3. Provide default values for every option except `steps_per_epoch`.
         """
-        callbacks = (callbacks or []) + DEFAULT_CALLBACKS()
-        monitors = (monitors or []) + DEFAULT_MONITORS()
+        assert _sentinel is None, "Please call `train_with_defaults` with keyword arguments only!"
+        callbacks = copy.copy(callbacks or [])
+        monitors = DEFAULT_MONITORS() if monitors is None else monitors
+        extra_callbacks = DEFAULT_CALLBACKS() if extra_callbacks is None else extra_callbacks
+        callbacks.extend(extra_callbacks)
 
         assert steps_per_epoch is not None
         session_creator = session_creator or NewSessionCreator()

tensorpack/train/config.py

@@ -12,7 +12,7 @@ from ..callbacks import (
 from ..dataflow.base import DataFlow
 from ..graph_builder.model_desc import ModelDescBase
 from ..utils import logger
-from ..tfutils.sessinit import JustCurrentSession, SessionInit, SaverRestore
+from ..tfutils.sessinit import SessionInit, SaverRestore
 from ..tfutils.sesscreate import NewSessionCreator
 from ..input_source import InputSource
 
@@ -52,21 +52,20 @@ def DEFAULT_MONITORS():
 
 class TrainConfig(object):
     """
-    A collection of options to be used for trainers.
+    A collection of options to be used for single-cost trainers.
     """
 
     def __init__(self,
-                 dataflow=None, data=None, model=None,
+                 dataflow=None, data=None,
+                 model=None,
                  callbacks=None, extra_callbacks=None, monitors=None,
                  session_creator=None, session_config=None, session_init=None,
-                 starting_epoch=1, steps_per_epoch=None, max_epoch=99999,
-                 nr_tower=1, tower=None,
-                 **kwargs):
+                 starting_epoch=1, steps_per_epoch=None, max_epoch=99999):
         """
         Args:
             dataflow (DataFlow):
             data (InputSource):
-            model (ModelDescBase):
+            model (ModelDesc):
 
             callbacks (list): a list of :class:`Callback` to perform during training.
             extra_callbacks (list): the same as ``callbacks``. This argument
@@ -107,20 +106,18 @@ class TrainConfig(object):
             assert_type(model, ModelDescBase)
         self.model = model
 
-        if callbacks is None:
-            callbacks = []
+        if callbacks is not None:
             assert_type(callbacks, list)
+        self.callbacks = callbacks
         if extra_callbacks is not None:
-            self._callbacks = callbacks + extra_callbacks
-        else:
-            self._callbacks = callbacks + DEFAULT_CALLBACKS()
-
-        self.monitors = monitors if monitors is not None else DEFAULT_MONITORS()
-
-        if session_init is None:
-            session_init = JustCurrentSession()
+            assert_type(extra_callbacks, list)
+        self.extra_callbacks = extra_callbacks
+        if monitors is not None:
+            assert_type(monitors, list)
+        self.monitors = monitors
+        if session_init is not None:
+            assert_type(session_init, SessionInit)
         self.session_init = session_init
-        assert_type(self.session_init, SessionInit)
 
         if session_creator is None:
             if session_config is not None:
@@ -149,27 +146,6 @@ class TrainConfig(object):
         self.starting_epoch = int(starting_epoch)
         self.max_epoch = int(max_epoch)
 
-        # Tower stuff are for Trainer v1 only:
-        nr_tower = max(nr_tower, 1)
-        self.nr_tower = nr_tower
-        if tower is not None:
-            assert self.nr_tower == 1, "Cannot set both nr_tower and tower in TrainConfig!"
-            self.tower = tower
-
-        assert len(kwargs) == 0, 'Unknown arguments: {}'.format(str(kwargs.keys()))
-
-    @property
-    def nr_tower(self):
-        return len(self.tower)
-
-    @nr_tower.setter
-    def nr_tower(self, value):
-        self.tower = list(range(value))
-
-    @property
-    def callbacks(self):        # disable setter
-        return self._callbacks
-
 
 class AutoResumeTrainConfig(TrainConfig):
     """

tensorpack/train/interface.py

@@ -76,14 +76,16 @@ def launch_train_with_config(config, trainer):
     inputs_desc = model.get_inputs_desc()
     input = config.data or config.dataflow
     input = apply_default_prefetch(input, trainer)
-    if config.nr_tower > 1:
-        logger.warn("With trainer v2, setting tower in TrainConfig has no effect.")
-        logger.warn("It's enough to set the tower when initializing the trainer.")
 
     trainer.setup_graph(
         inputs_desc, input,
         model._build_graph_get_cost, model.get_optimizer)
-    trainer.train(
-        config.callbacks, config.monitors,
-        config.session_creator, config.session_init,
-        config.steps_per_epoch, config.starting_epoch, config.max_epoch)
+    trainer.train_with_defaults(
+        callbacks=config.callbacks,
+        monitors=config.monitors,
+        session_creator=config.session_creator,
+        session_init=config.session_init,
+        steps_per_epoch=config.steps_per_epoch,
+        starting_epoch=config.starting_epoch,
+        max_epoch=config.max_epoch,
+        extra_callbacks=config.extra_callbacks)