update docs

docs/conf.py

@@ -363,6 +363,7 @@ _DEPRECATED_NAMES = set([
     'TFRecordData',
     'dump_dataflow_to_lmdb',
     'dump_dataflow_to_tfrecord',
+    'pyplot2img',
 
     # renamed stuff:
     'DumpTensor',

docs/requirements.txt

@@ -6,4 +6,5 @@ Sphinx>=1.6
 recommonmark==0.4.0
 sphinx_rtd_theme
 mock
+matplotlib
 tensorflow==1.4.0

tensorpack/graph_builder/__init__.py

@@ -28,7 +28,7 @@ def global_import(name):
 
 
 _CURR_DIR = os.path.dirname(__file__)
-_SKIP = []
+_SKIP = ['training', 'distributed']
 for _, module_name, _ in iter_modules(
         [_CURR_DIR]):
     srcpath = os.path.join(_CURR_DIR, module_name + '.py')

tensorpack/input_source/input_source.py

@@ -40,6 +40,9 @@ class PlaceholderInput(InputSource):
     """
     Just produce placeholders as input tensors.
     """
+    def __init__(self):
+        pass
+
     def _setup(self, inputs):
         self._all_placehdrs = [v.build_placeholder_reuse() for v in inputs]
 

tensorpack/input_source/input_source_base.py

@@ -75,8 +75,8 @@ class InputSource(object):
         """
         Returns:
             list: A list of tensors corresponding to the inputs of the model,
-                used as input of :func:`build_graph`.
-                For non-placeholder tensors, should always create and return new tensors when called.
+            used as input of :func:`build_graph`.
+            For non-placeholder tensors, should always create and return new tensors when called.
         """
         return self._get_input_tensors()
 
@@ -92,7 +92,7 @@ class InputSource(object):
 
         Returns:
             list[Callback]: extra callbacks needed by this InputSource.
-                callbacks of InputSource cannot use any `trigger*()` method.
+            callbacks of InputSource cannot use any `trigger*()` method.
         """
         self._setup(inputs_desc)
         self._setup_done = True

tensorpack/tfutils/gradproc.py

@@ -18,7 +18,10 @@ __all__ = ['GradientProcessor',
 
 @six.add_metaclass(ABCMeta)
 class GradientProcessor(object):
-    """ Base class for all gradient processors.
+    """
+    Base class for all gradient processors.
+    Gradient processors can be applied to optimizers by
+    :func:`optimizer.apply_grad_processors`.
 
     Subclass should override the ``_process()`` method.
     """

tensorpack/tfutils/optimizer.py

@@ -4,6 +4,8 @@
 
 import tensorflow as tf
 from contextlib import contextmanager
+
+from ..utils.develop import HIDE_DOC
 from .gradproc import FilterNoneGrad, GradientProcessor
 
 __all__ = ['apply_grad_processors', 'ProxyOptimizer',
@@ -20,15 +22,19 @@ class ProxyOptimizer(tf.train.Optimizer):
         super(ProxyOptimizer, self).__init__(False, name)
         self._opt = opt
 
+    @HIDE_DOC
     def compute_gradients(self, *args, **kwargs):
         return self._opt.compute_gradients(*args, **kwargs)
 
+    @HIDE_DOC
     def get_slot(self, *args, **kwargs):
         return self._opt.get_slot(*args, **kwargs)
 
+    @HIDE_DOC
     def get_slot_names(self, *args, **kwargs):
         return self._opt.get_slot_names(*args, **kwargs)
 
+    @HIDE_DOC
     def apply_gradients(self, *args, **kwargs):
         return self._opt.apply_gradients(*args, **kwargs)
 
@@ -85,6 +91,7 @@ class PostProcessOptimizer(ProxyOptimizer):
         self._func = func
         self._colocate = colocate
 
+    @HIDE_DOC
     def apply_gradients(self, grads_and_vars, global_step=None, name=None):
         update_op = super(PostProcessOptimizer, self).apply_gradients(
             grads_and_vars, global_step)
@@ -131,8 +138,8 @@ class VariableAssignmentOptimizer(PostProcessOptimizer):
 
 class AccumGradOptimizer(ProxyOptimizer):
     """
-    An optimizer which accumulates gradients across :math:`k` :meth:`minimize` calls,
-    and apply them together in every :math:`k`th :meth:`minimize` call.
+    An optimizer which accumulates gradients across :math:`k` :meth:`minimize` executions,
+    and apply them together in every :math:`k` th :meth:`minimize` execution.
     This is roughly the same as using a :math:`k` times larger batch size plus a
     :math:`k` times larger learning rate, but uses much less memory.
 
@@ -157,6 +164,7 @@ class AccumGradOptimizer(ProxyOptimizer):
             slots.append(s)
         return slots
 
+    @HIDE_DOC
     def apply_gradients(self, grads_and_vars, global_step=None, name=None):
         assert global_step is None, \
             "AccumGradOptimizer doesn't support the option global_step! " \

tensorpack/tfutils/tower.py

@@ -29,6 +29,9 @@ class BaseTowerContext(object):
     @HIDE_DOC
     def __init__(self, ns_name, vs_name=''):
         """
+        This is not supposed to be used by users.
+        You need to use :func:`TowerContext` to create a :class:`BaseTowerContext`.
+
         Args:
             ns_name (str): The name scope of the tower.
             vs_name (str): Open a new variable scope with this name.
@@ -214,7 +217,7 @@ class PredictTowerContext(BaseTowerContext):
 
 def get_current_tower_context():
     """
-    When called inside a TowerContext, return the TowerContext.
+    When called inside a TowerContext, returns the TowerContext.
 
     Returns:
         a :class:`BaseTowerContext` instance.

tests/dev/git-hooks/pre-commit

@@ -7,7 +7,7 @@ GIT_ARG="--git-dir ../.git --work-tree .."
 # find out modified python files, so that we ignored unstaged files
 # exclude ../docs
 MOD=$(git $GIT_ARG status -s \
-	| grep -E '\.py$' | grep -v '../docs' \
+	| grep -E '\.py$' | grep -v '../docs' | grep -v '__init__' \
 	| grep -E '^ *M|^ *A' | cut -c 4- )
 if [[ -n $MOD ]]; then
 	flake8 $MOD