bindings/python/doc: fix most Sphinx warnings and some misc. errors

bindings/python/cntk/debugging/debug.py

@@ -89,7 +89,7 @@ def save_as_legacy_model(root_op, filename):
     For debugging purposes only, very likely to be deprecated in the future.
 
     Args:
-        root_op (:class:`~cntk.functions.Function`): op of the graph to save
+        root_op (:class:`~cntk.ops.functions.Function`): op of the graph to save
         filename (str): filename to store the model in.
     '''
     cntk_py.save_as_legacy_model(root_op, filename)

bindings/python/cntk/initializer.py

@@ -18,7 +18,7 @@ def uniform(scale, seed=None):
         seed (int): random seed
 
     Returns:
-        initializer for :class:`cntk.variables.Parameter`
+        initializer for :class:`cntk.ops.variables.Parameter`
         initialized to uniform distribution between `scale*[-1.0, 1.0]`
         note this maps to the "uniform1" distribution in BrainScript. 
     '''
@@ -38,7 +38,7 @@ def normal(scale, output_rank=SentinelValueForInferParamInitRank, filter_rank=Se
         seed (int): random seed
 
     Returns:
-        initializer for :class:`cntk.variables.Parameter`
+        initializer for :class:`cntk.ops.variables.Parameter`
         initialized to normal distribution with mean `0` and standard deviation `scale`. 
     '''
     if seed is None:

bindings/python/cntk/internal/sanitize.py

@@ -157,7 +157,7 @@ def sanitize_batch(var, batch, seq_starts=None, device=None):
 
 def sanitize_value(shape, value, dtype, device):
     '''
-    Converts a given ``value`` to an :class:`~cntk.NDArrayView` object that can be passed to
+    Converts a given ``value`` to an :class:`~cntk.core.NDArrayView` object that can be passed to
     the CNTK core.
 
     Args:
@@ -169,7 +169,7 @@ def sanitize_value(shape, value, dtype, device):
          on
 
     Returns:
-        :class:`~cntk.NDArrayView` object representing ``value``
+        :class:`~cntk.core.NDArrayView` object representing ``value``
     '''
     from .. import NDArrayView
     if value is None:
@@ -216,8 +216,8 @@ def sanitize_var_map(op_arguments, arguments, precision=None,
     handed off to the evaluation methods
     (:meth:`~cntk.ops.functions.Function.forward`,
     :meth:`~cntk.ops.functions.Function.backward`,
-    :meth:`~cntk.Trainer.train_minibatch` and
-    :meth:`~cntk.Trainer.test_minibatch`).
+    :meth:`~cntk.train.trainer.Trainer.train_minibatch` and
+    :meth:`~cntk.train.trainer.Trainer.test_minibatch`).
 
     Args:
         op_arguments (:class:`~cntk.ops.functions.Function`): arguments of the
@@ -264,7 +264,7 @@ def sanitize_var_map(op_arguments, arguments, precision=None,
          map are converted to the underlying value (:class:`~cntk.core.Value`)
          instances (default), or if they should remain intact, as they contain
          additional meta information required by the Trainer (specifically, by
-         the :meth:`~cntk.Trainer.train_minibatch` method).
+         the :meth:`~cntk.train.trainer.Trainer.train_minibatch` method).
 
     Returns:
         `dict` that maps variables to sanitized batches

bindings/python/cntk/io/__init__.py

@@ -84,7 +84,7 @@ class MinibatchSource(cntk_py.MinibatchSource):
     A `MinibatchSource` can be indexed by the stream name, which will return a
     Parent class of all minibatch sources.  A `MinibatchSource` can be indexed by the stream name, which will return a
     :class:`MinibatchData` object that can be passed e.g. to the
-    :func:`~cntk.trainer.Trainer.train_minibatch` function.
+    :func:`~cntk.train.trainer.Trainer.train_minibatch` function.
 
     Args:
         deserializers (`list`, defaults to empty): list of deserializers
@@ -366,7 +366,7 @@ def minibatch_source(self):
         '''
         Creates an instance of :class:`MinibatchSource` from this
         instance, which can be used to feed data into the `eval()` methods of
-        the graph nodes or the `train_minibatch()` of :class:`~cntk.trainer.Trainer`.
+        the graph nodes or the `train_minibatch()` of :class:`~cntk.train.trainer.Trainer`.
 
         Returns:
             cntk.io.MinibatchSource:

bindings/python/cntk/learners/__init__.py

@@ -428,7 +428,7 @@ def sgd(parameters, lr,
          with truncation
 
     Returns:
-        Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
+        Instance of a :class:`~cntk.learners.Learner` that can be passed to the :class:`~cntk.train.trainer.Trainer`
 
     See also:
         [1] L. Bottou. `Stochastic Gradient Descent Tricks
@@ -479,8 +479,8 @@ def momentum_sgd(parameters, lr, momentum, unit_gain=default_unit_gain_value(),
          with truncation
 
     Returns:
-        Instance of a :class:`~cntk.learner.Learner` that can be passed to the
-        :class:`~cntk.trainer.Trainer`
+        Instance of a :class:`~cntk.learners.Learner` that can be passed to the
+        :class:`~cntk.train.trainer.Trainer`
     '''
     _verify_learning_rate_type(lr)
     _verify_momentum_type(momentum)
@@ -530,8 +530,8 @@ def nesterov(parameters, lr, momentum, unit_gain=default_unit_gain_value(),
          with truncation
 
     Returns:
-        Instance of a :class:`~cntk.learner.Learner` that can be passed to the
-        :class:`~cntk.trainer.Trainer`.
+        Instance of a :class:`~cntk.learners.Learner` that can be passed to the
+        :class:`~cntk.train.trainer.Trainer`.
 
     See also:
         [1] Y. Nesterov. A Method of Solving a Convex Programming Problem with Convergence Rate O(1/ sqrt(k)). Soviet Mathematics Doklady, 1983.
@@ -585,7 +585,7 @@ def adagrad(parameters, lr, need_ave_multiplier=True,
          with truncation
 
     Returns:
-        Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
+        Instance of a :class:`~cntk.learners.Learner` that can be passed to the :class:`~cntk.train.trainer.Trainer`
 
     See also:
         [1]  J. Duchi, E. Hazan, and Y. Singer. `Adaptive Subgradient Methods
@@ -641,7 +641,7 @@ def fsadagrad(parameters, lr, momentum, unit_gain=default_unit_gain_value(),
          with truncation
 
     Returns:
-        Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
+        Instance of a :class:`~cntk.learners.Learner` that can be passed to the :class:`~cntk.train.trainer.Trainer`
 
     '''
     _verify_learning_rate_type(lr)
@@ -695,7 +695,7 @@ def adam(parameters, lr, momentum, unit_gain=default_unit_gain_value(),
          with truncation
 
     Returns:
-        Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
+        Instance of a :class:`~cntk.learners.Learner` that can be passed to the :class:`~cntk.train.trainer.Trainer`
 
     See also:
         [1] D. Kingma, J. Ba. `Adam: A Method for Stochastic Optimization
@@ -757,7 +757,7 @@ def adam_sgd(parameters, lr, momentum, unit_gain=default_unit_gain_value(),
          with truncation
 
     Returns:
-        Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
+        Instance of a :class:`~cntk.learners.Learner` that can be passed to the :class:`~cntk.train.trainer.Trainer`
 
     See also:
         [1] D. Kingma, J. Ba. `Adam: A Method for Stochastic Optimization
@@ -814,7 +814,7 @@ def rmsprop(parameters, lr,
          with truncation
 
     Returns:
-        Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
+        Instance of a :class:`~cntk.learners.Learner` that can be passed to the :class:`~cntk.train.trainer.Trainer`
     '''
     _verify_learning_rate_type(lr)
     gaussian_noise_injection_std_dev = \

bindings/python/cntk/logging/progress_print.py

@@ -302,7 +302,7 @@ def update_with_trainer(self, trainer, with_metric=False):
         ``trainer``.
 
         Args:
-            trainer (:class:`cntk.trainer.Trainer`): trainer from which information is gathered
+            trainer (:class:`cntk.train.trainer.Trainer`): trainer from which information is gathered
             with_metric (`bool`): whether to update the metric accumulators
         '''
         if self.total_updates == 0:
@@ -421,7 +421,7 @@ class TensorBoardProgressWriter(cntk_py.ProgressWriter):
         log_dir (`string`, default '.'): directory where to create a TensorBoard event file.
         rank (`int` or `None`, default `None`): rank of a worker when using distributed training, or `None` if
          training locally. If not `None`, event files will be created only by rank 0.
-        model (:class:`cntk.ops.Function` or `None`, default `None`): model graph to plot.
+        model (:class:`cntk.ops.functions.Function` or `None`, default `None`): model graph to plot.
     '''
 
     def __init__(self, freq=None, log_dir='.', rank=None, model=None):

bindings/python/cntk/ops/functions.py

@@ -1221,7 +1221,7 @@ def infer_outputs(self):
         outputs.
 
         Output variables are created by
-        :meth:`~cntk.ops.functions.output_variable`.
+        :meth:`~cntk.ops.output_variable`.
         '''
         raise NotImplementedError('infer_outputs has to be overwritten')
 

bindings/python/cntk/train/training_session.py

@@ -11,7 +11,7 @@
 from ..io import _py_dict_to_cntk_dict
 
 __doc__ = '''\
-A training session encapsulates a typical training loop and binds together a minibatch source that is used for training, a :doc:`trainer <cntk.trainer>` and an optional cross validation minibatch source. A training session takes care of consistent checkpointing and progress printing with specified frequencies. 
+A training session encapsulates a typical training loop and binds together a minibatch source that is used for training, a :doc:`trainer <cntk.train.trainer>` and an optional cross validation minibatch source. A training session takes care of consistent checkpointing and progress printing with specified frequencies. 
 '''
 class CheckpointConfig(cntk_py.CheckpointConfig):
     '''
@@ -94,14 +94,14 @@ def __init__(self, source=None, mb_size=None, frequency=None, callback=None):
 
 class TrainingSession(cntk_py.TrainingSession):
     '''
-    The instance of the class should be created by using :func:`~cntk.training_session.training_session` function.
+    The instance of the class should be created by using :func:`~cntk.train.training_session.training_session` function.
 
     A training session trains a model using the specified ``trainer`` and configs.
     Different aspects of training such as data sources, checkpointing, cross validation, progress printing
     can be configured using the corresponding config classes.
 
     Args:
-        trainer (:class:`~cntk.trainer.Trainer`): trainer
+        trainer (:class:`~cntk.train.trainer.Trainer`): trainer
         mb_source (:class:`~cntk.io.MinibatchSource`): minibatch source used for training
         mb_size (:class:`~cntk.cntk_py.minibatch_size_schedule` or int): minibatch size schedule for training
         var_to_stream (dict): mapping between input variables and input streams
@@ -256,7 +256,7 @@ def training_session(training_minibatch_source=None,        # deprecated, will b
 
     Args: 
         training_minibatch_source (:class:`~cntk.io.MinibatchSource`): !DEPRECATED! use mb_source instead
-        trainer (:class:`~cntk.trainer.Trainer`): trainer
+        trainer (:class:`~cntk.train.trainer.Trainer`): trainer
         mb_size_schedule (:class:`~cntk.cntk_py.minibatch_size_schedule`): !DEPRECATED! use mb_size instead
         progress_printer (list): !DEPRECATED! list of progress writers from :mod:`cntk.utils`
         model_inputs_to_mb_source_mapping (dict): !DEPRECATED! use var_to_stream instead

bindings/python/cntk/utils/__init__.py

@@ -130,7 +130,7 @@ def get_train_loss(trainer):
     Fetch the train loss from the last minibatch and copy it to the CPU in case it is on the GPU.
 
     Args:
-        trainer (:class:`~cntk.trainer.Trainer`): the trainer used.
+        trainer (:class:`~cntk.train.trainer.Trainer`): the trainer used.
     Returns:
         the loss value
     '''

bindings/python/doc/extend.rst

@@ -146,8 +146,8 @@ debugging mode and can start inspection.
 User learners
 -------------
 Implementing a custom learner in pure Python is accomplished by
- - creating a class that inherits from :class:`cntk.learner.UserLearner`
- - implementing its :meth:`~cntk.learner.UserLearner.update` method
+ - creating a class that inherits from :class:`cntk.learners.UserLearner`
+ - implementing its :meth:`~cntk.learners.UserLearner.update` method
 
 Here is an example, how normal stochastic gradient descent would be
 reimplemented in a naive way::

bindings/python/doc/layerref.rst

@@ -1,14 +1,14 @@
 Layers Library Reference
 ========================
 
-Note: This documentation has not yet been completely updated w.r.t. the latest update of the Layers library.
-It should be correct but miss several new options and layer types.
+Note: This documentation has not yet been completely updated with respect to the latest update of the Layers library.
+It should be correct but misses several new options and layer types.
 
 CNTK predefines a number of common "layers," which makes it very easy to
 write simple networks that consist of standard layers layered on top of
 each other. Layers are function objects that can be used like a regular
-:class:`~cntk.ops.functions.Function` but hold learnable parameters 
-and have an additional pair of ``()`` to pass construction parameters 
+:class:`~cntk.ops.functions.Function` but hold learnable parameters
+and have an additional pair of ``()`` to pass construction parameters
 or attributes.
 
 For example, this is the network description for a simple 1-hidden layer
@@ -43,7 +43,7 @@ and invoke it like this:
 
     p = my_model(features)
 
-Built on top of ``Sequential()`` is :ref:`layerstack`,
+Built on top of ``Sequential()`` is :ref:`for`,
 which allows to easily create models with repetitions. For example, a
 2011-style feed-forward speech-recognition network with 6 hidden sigmoid
 layers of identical dimensions can be written like this:
@@ -82,13 +82,13 @@ to define scopes with locally changed defaults, using one of the following two f
     with default_options_for(FUNCTION, OPT1=VAL1, OPT2=VAL2, ...):
         # scope with modified defaults for FUNCTION only
 
-The following options can be overridden with the ``with`` statement: 
+The following options can be overridden with the ``with`` statement:
 
-    - ``init`` (default: ``glorot_uniform()``): initializer specification, for :ref:`dense`, :ref:`convolution`, and :ref:`embedding` 
-    - ``activation`` (default: ``None``): activation function, for ``Dense()`` and ``Convolution()`` 
+    - ``init`` (default: ``glorot_uniform()``): initializer specification, for :ref:`dense`, :ref:`convolution`, and :ref:`embedding`
+    - ``activation`` (default: ``None``): activation function, for ``Dense()`` and ``Convolution()``
     - ``bias`` (default: ``True``): have a bias, for ``Dense()`` and ``Convolution()``
-    - ``init_bias`` (default: ``0``): initializer specification for the bias, for ``Dense()`` and ``Convolution()`` 
-    - ``initial_state`` (default: ``None``): initial state to use in ``Recurrence()`` :ref:`recurrence` 
+    - ``init_bias`` (default: ``0``): initializer specification for the bias, for ``Dense()`` and ``Convolution()``
+    - ``initial_state`` (default: ``None``): initial state to use in ``Recurrence()`` :ref:`recurrence`
     - ``use_peepholes`` (default: ``False``): use peephole connections in ``LSTM()`` :ref:`lstm`
 
 The second for allows to set default options on a
@@ -242,7 +242,7 @@ Description
 Use these factory functions to create a fully-connected layer. It
 creates a function object that contains a learnable weight matrix and,
 unless ``bias=False``, a learnable bias. The function object can be used
-like a function, which implements one of these formulas (using Python 
+like a function, which implements one of these formulas (using Python
 3.5 ``@`` operator for matrix multiplication):
 
 ::
@@ -515,12 +515,12 @@ Description
 ~~~~~~~~~~~
 
 Use this factory function to create a global pooling operation. Use
-``GlobalMaxPooling()`` to compute the maximum over all spatial data, 
+``GlobalMaxPooling()`` to compute the maximum over all spatial data,
 or ``GlobalAveragePooling()`` to take their average.
 
-The global pooling operation infer the pooling window shape from the input 
+The global pooling operation infer the pooling window shape from the input
 tensor and create a pooling function with pooling window size that
-matches the input spatial dimension. It then computes either the 
+matches the input spatial dimension. It then computes either the
 maximum or the average of all the values inside the inferred pooling
 window.
 
@@ -529,7 +529,7 @@ Example:
 
 ::
 
-    p = GlobalMaxPooling()(c)    
+    p = GlobalMaxPooling()(c)
 
 Dropout()
 ------------------------------
@@ -544,7 +544,7 @@ Parameters
 ~~~~~~~~~~
 
 -  ``dropout_rate``: a fraction between [0, 1) that specifies the probability by which
-   the dropout operation will randomly set elements of the input to zero. 0 mean 
+   the dropout operation will randomly set elements of the input to zero. 0 mean
    select everything and close to 1 mean drop every element.
 
 Return Value
@@ -555,15 +555,15 @@ A function that implements the desired dropout layer.
 Description
 ~~~~~~~~~~~
 
-Use this factory function to create a dropout operation with a specific 
+Use this factory function to create a dropout operation with a specific
 dropout rate.
 
 Example:
 ~~~~~~~~
 
 ::
 
-    p = Dropout(0.5)(c)    
+    p = Dropout(0.5)(c)
 
 .. _embedding:
 
@@ -1035,7 +1035,7 @@ The ``FGH`` function defined above means the same as
 
 ::
 
-    y = H(G(F(x))) 
+    y = H(G(F(x)))
 
 This is known as `function
 composition <https://en.wikipedia.org/wiki/Function_composition>`_,
@@ -1081,10 +1081,10 @@ deep-neural network work on speech recognition:
 
     my_model = Sequential ([
         Dense(2048, activation=sigmoid),  # four hidden layers
-        Dense(2048, activation=sigmoid), 
-        Dense(2048, activation=sigmoid), 
-        Dense(2048, activation=sigmoid), 
-        Dense(9000, activation=softmax)   # note: last layer is a softmax 
+        Dense(2048, activation=sigmoid),
+        Dense(2048, activation=sigmoid),
+        Dense(2048, activation=sigmoid),
+        Dense(9000, activation=softmax)   # note: last layer is a softmax
     )
     features = Input(40)
     p = my_model(features)