binding/python: more doc fixes

bindings/python/cntk/core.py

@@ -84,7 +84,7 @@ def from_csr(csr_array, device=None, read_only=False, borrow=False):
         format.
 
         Args:
-            csr_array (scipy.sparse.csr.csr_matrix): SciPy sparse matrix in CSR
+            csr_array (scipy.sparse.csr_matrix): SciPy sparse matrix in CSR
              format
             device (:class:`~cntk.device.DeviceDescriptor`): device this value
              should be put on
@@ -115,7 +115,7 @@ def from_data(data, device=None, read_only=False, borrow=False):
         array in CSR format.
 
         Args:
-            data (numpy.ndarray or scipy.sparse.csr.csr_matrix): data
+            data (numpy.ndarray or scipy.sparse.csr_matrix): data
             device (:class:`~cntk.device.DeviceDescriptor`): device this value
              should be put on
             read_only (bool, optional): whether the data can be modified or
@@ -397,7 +397,7 @@ def one_hot(batch, num_classes, dtype=None, device=None):
 
         Args:
             batch (list of lists of integers): batch input data of indices
-            sample_shape (integer or tuple): number of classes or shape of each sample whose trailing axis is one_hot
+            sample_shape (int or tuple): number of classes or shape of each sample whose trailing axis is one_hot
             dtype (`np.float32`, `np.float64`, default None): data type
             device (:class:`~cntk.device.DeviceDescriptor`, default None): device
              this value should be put on

bindings/python/cntk/debugging/debug.py

@@ -18,8 +18,8 @@
         n - execute the next node
         n <number> - execute the next <number> nodes
 
-        u f - exeucte until forward pass (like 'n' when already in forward pass)
-        u b - exeucte until backward pass (like 'n' when already in backward pass)
+        u f - execute until forward pass (like 'n' when already in forward pass)
+        u b - execute until backward pass (like 'n' when already in backward pass)
         u name - execute until a node with that name is hit
         u <lambda> - execute until the lambda expression is True. Examples:
                      Until a Times node is hit:
@@ -29,7 +29,7 @@
                      Until the variance of the input exceeds 1 (np = numpy):
                          lambda arg, node: np.var(arg) > 1
 
-        c - exeucte until end
+        c - execute until end
         p - print input (forward) or root gradients (backward)
         d - drop into a pdb shell
         q - quit\
@@ -112,18 +112,23 @@ def set_computation_network_track_gap_nans(enable):
     For debugging purposes only.
 
     Args:
-        enable (Boolean): whether to enable gap nans tracking (with performance impact)
+        enable (bool): whether to enable gap nans tracking (with performance impact)
     '''
     cntk_py.set_computation_network_track_gap_nans(enable)
 
 
 def set_computation_network_trace_level(level):
     '''
     Set trace level to the computation network. Currently supported values:
-       0        : turn off trace
-       1        : output nodes' dimensions and some other static info
-       1000     : output each node's abs sum of elements in its value matrix for every forward/backward
-       1000000  : output each node's full matrix for every forward/backward
+
+       0
+         turn off trace
+       1
+         output nodes' dimensions and some other static info
+       1000
+         output each node's abs sum of elements in its value matrix for every forward/backward
+       1000000
+         output each node's full matrix for every forward/backward
 
     Args:
         level (int): trace level
@@ -391,6 +396,8 @@ def node_filter(x):
 def debug_model(model, in_stream=sys.stdin, out_stream=sys.stdout,
                 exit_func=sys.exit):
     '''
+    debug_model(model, in_stream=sys.stdin, out_stream=sys.stdout, exit_func=sys.exit)
+
     Returns a cloned model that has debug nodes inserted everywhere. When the
     graph is evaluated or trained, those nodes will allow to inspect the graph.
 

bindings/python/cntk/internal/sanitize.py

@@ -124,7 +124,7 @@ def sanitize_batch(var, batch, seq_starts=None, device=None):
 
            * a single NumPy array denoting the full minibatch
            * a list of NumPy arrays or SciPy sparse CSR matrices each representing a sequence
-           * a :class:`~cntk.core.Value` object (e.g. returned by :func:`one_hot`)
+           * a :class:`~cntk.core.Value` object (e.g. returned by :func:`cntk.core.Value.one_hot`)
         seq_starts (list of bools or None): if None, every sequence is
          treated as a new sequence. Otherwise, it is interpreted as a list of
          Booleans one for each sequence in the batch that tell whether a

bindings/python/cntk/io/__init__.py

@@ -13,6 +13,8 @@
 import uuid
 
 INFINITELY_REPEAT = cntk_py.MinibatchSource.infinitely_repeat
+'''int: constant used to specify a minibatch scheduling unit to equal the size of the full data sweep.'''
+
 FULL_DATA_SWEEP = cntk_py.MinibatchSource.full_data_sweep
 INFINITE_SAMPLES = cntk_py.MinibatchSource.infinite_samples
 DEFAULT_RANDOMIZATION_WINDOW = cntk_py.MinibatchSource.default_randomization_window
@@ -103,8 +105,7 @@ class MinibatchSource(cntk_py.MinibatchSource):
           the entire data sweep (as indicated by the MinibatchSource) and parameters
           change their values on the sweep-by-sweep basis specified by the schedule.
           **Important:**
-          Click `here <https://github.com/Microsoft/CNTK/wiki/BrainScript-epochSize-and-Python-epoch_size-in-CNTK>`_ for a full description of this parameter.
-        distributed_after (int, defaults to cntk.io.INFINITE_SAMPLES): sample count after which minibatch source becomes distributed
+          Click `here <https://github.com/Microsoft/CNTK/wiki/BrainScript-epochSize-and-Python-epoch_size-in-CNTK>`__ for a full description of this parameter.
         multithreaded_deserializer (`bool`, defaults to `None`): using multi threaded deserializer
         frame_mode (`bool`, defaults to `False`): Specifies if data should be randomized and returned at the frame
          or sequence level. When  true , input sequence are split into frames.
@@ -186,7 +187,7 @@ def next_minibatch(self, minibatch_size_in_samples,
             minibatch_size_in_samples (int): number of samples to retrieve for
               the next minibatch. Must be > 0.
               **Important:**
-              Click `here <https://github.com/Microsoft/CNTK/wiki/BrainScript-minibatchSize-and-Python-minibatch_size_in_samples-in-CNTK>`_ for a full description of this parameter. 
+              Click `here <https://github.com/Microsoft/CNTK/wiki/BrainScript-minibatchSize-and-Python-minibatch_size_in_samples-in-CNTK>`__ for a full description of this parameter. 
             input_map (dict): mapping of :class:`~cntk.ops.variables.Variable`
               to :class:`~cntk.cntk_py.StreamInformation` which will be used to convert the
               returned data.
@@ -236,7 +237,7 @@ def restore_from_checkpoint(self, checkpoint):
         Restores the MinibatchSource state from the specified checkpoint.
 
         Args:
-            checkpoint (:class:`~cntk_py.Dictionary`): checkpoint to restore from
+            checkpoint (:class:`~cntk.cntk_py.Dictionary`): checkpoint to restore from
         '''
         super(MinibatchSource, self).restore_from_checkpoint(checkpoint)
 
@@ -325,7 +326,9 @@ class _ReaderConfig(dict):
           samples. If no `epoch_size` is provided, this parameter is substituted
           by the size of the full data sweep with infinite repeat, in which case the scheduling unit is
           the entire data sweep (as indicated by the MinibatchSource) and parameters
-          change their values on the sweep-by-sweep basis specified by the schedule. **Important:** `click here <https://github.com/Microsoft/CNTK/wiki/BrainScript-epochSize-and-Python-epoch_size-in-CNTK>`_ for a full description of this parameter. 
+          change their values on the sweep-by-sweep basis specified by the schedule.
+          **Important:**
+          Click `here <https://github.com/Microsoft/CNTK/wiki/BrainScript-epochSize-and-Python-epoch_size-in-CNTK>`__ for a full description of this parameter. 
         distributed_after (int, defaults to `cntk.io.INFINITE_SAMPLES`): sample count after which reader becomes distributed
         multithreaded_deserializer (`bool`, defaults to `None`): using multi threaded deserializer
         frame_mode (`bool`, defaults to `False`): Specifies if data should be randomized and returned at the frame

bindings/python/cntk/layers/layers.py

@@ -25,7 +25,7 @@ def Dense(shape, activation=default_override_or(identity), init=default_override
           bias=default_override_or(True), init_bias=default_override_or(0),
           name=''):
     '''
-    Dense(shape, activation=identity, init=glorot_uniform(),input_rank=None, map_rank=None, bias=True, init_bias=0, name='')
+    Dense(shape, activation=identity, init=glorot_uniform(), input_rank=None, map_rank=None, bias=True, init_bias=0, name='')
 
     Layer factory function to create an instance of a fully-connected linear layer of the form
     `activation(input @ W + b)` with weights `W` and bias `b`, and `activation` and `b` being optional.
@@ -114,7 +114,7 @@ def dense(x):
 
 def Embedding(shape=None, init=default_override_or(glorot_uniform()), weights=None, name=''):
     '''
-    Embedding(shape=None, init=glorot_uniform(), weights=None, name=''):
+    Embedding(shape=None, init=glorot_uniform(), weights=None, name='')
 
     Layer factory function to create a embedding layer.
 
@@ -263,7 +263,7 @@ def Convolution(filter_shape,     # shape of receptive field, e.g. (3,3)
                 max_temp_mem_size_in_samples=0,
                 op_name='Convolution', name=''):
     '''
-    Convolution(filter_shape, num_filters=None, sequential=False, activation=identity, init=glorot_uniform(), pad=False, strides=1, sharing=True, bias=True, init_bias=0, reduction_rank=1, transpose=False, max_temp_mem_size_in_samples=0, op_name='Convolution', name='')
+    Convolution(filter_shape, num_filters=None, sequential=False, activation=identity, init=glorot_uniform(), pad=False, strides=1, sharing=True, bias=True, init_bias=0, reduction_rank=1, transpose_weight=False, max_temp_mem_size_in_samples=0, op_name='Convolution', name='')
 
     Layer factory function to create a convolution layer.
 
@@ -617,6 +617,8 @@ def ConvolutionTranspose(filter_shape,        # shape of receptive field, e.g. (
                          name=''):
 
     '''
+    ConvolutionTranspose(filter_shape, num_filters, activation=identity, init=glorot_uniform(), pad=False, strides=1, sharing=True, bias=True, init_bias=0, output_shape=None, reduction_rank=1, max_temp_mem_size_in_samples=0, name='')
+
     Layer factory function to create a convolution transpose layer.
 
     This implements a convolution_transpose operation over items arranged on an N-dimensional grid, such as pixels in an image.
@@ -739,6 +741,8 @@ def ConvolutionTranspose1D(filter_shape,        # a scalar, e.g., 3
                            output_shape=None, 
                            name=''):
     '''
+    ConvolutionTranspose1D(filter_shape, num_filters, activation=identity, init=glorot_uniform(), pad=False, strides=1, bias=True, init_bias=0, output_shape=None, name='')
+
     Layer factory function to create a 1D convolution transpose layer with optional non-linearity.
     Same as `ConvolutionTranspose()` except that filter_shape is verified to be 1-dimensional.
     See `ConvolutionTranspose()` for extensive documentation.
@@ -765,6 +769,8 @@ def ConvolutionTranspose2D(filter_shape,        # a 2D tuple, e.g., (3,3)
                            output_shape=None, 
                            name=''):
     '''
+    ConvolutionTranspose2D(filter_shape, num_filters, activation=identity, init=glorot_uniform(), pad=False, strides=1, bias=True, init_bias=0, output_shape=None, name='')
+
     Layer factory function to create a 2D convolution transpose layer with optional non-linearity.
     Same as `ConvolutionTranspose()` except that filter_shape is verified to be 2-dimensional.
     See `ConvolutionTranspose()` for extensive documentation.
@@ -792,6 +798,8 @@ def ConvolutionTranspose3D(filter_shape,        # a 3D tuple, e.g., (3,3,3)
                            output_shape=None, 
                            name=''):
     '''
+    ConvolutionTranspose3D(filter_shape, num_filters, activation=identity, init=glorot_uniform(), pad=False, strides=1, bias=True, init_bias=0, output_shape=None, name='')
+
     Layer factory function to create a 3D convolution transpose layer with optional non-linearity.
     Same as `ConvolutionTranspose()` except that filter_shape is verified to be 3-dimensional.
     See `ConvolutionTranspose()` for extensive documentation.

bindings/python/cntk/layers/models/attention.py

@@ -20,6 +20,8 @@ def AttentionModel(attention_dim, attention_span=None, attention_axis=None,
                    go_backwards=default_override_or(False),
                    enable_self_stabilization=default_override_or(True), name=''):
     '''
+    AttentionModel(attention_dim, attention_span=None, attention_axis=None, init=glorot_uniform(), go_backwards=False, enable_self_stabilization=True, name='')
+
     Layer factory function to create a function object that implements an attention model
     as described in Bahdanau, et al., "Neural machine translation by jointly learning to align and translate."
     '''

bindings/python/cntk/layers/sequence.py

@@ -164,6 +164,8 @@ def _sanitize_function(f):
 # TODO: allow to say sequential=False, axis=2, length=100, ... something like this
 def RecurrenceFrom(step_function, go_backwards=default_override_or(False), return_full_state=False, name=''):
     '''
+    RecurrenceFrom(step_function, go_backwards=False, return_full_state=False, name='')
+
     Layer factory function to create a function that runs a cell function recurrently over an input sequence, with initial state.
     This layer is very similar to :func:`~cntk.layers.sequence.Recurrence`,
     except that the initial state is data dependent, and thus passed to the layer function as a data input
@@ -275,6 +277,8 @@ def recurrence_from_3(h, c, a, x):
 # TODO: Can bidirectionality be an option of this? bidirectional=True?
 def Recurrence(step_function, go_backwards=default_override_or(False), initial_state=default_override_or(0), return_full_state=False, name=''):
     '''
+    Recurrence(step_function, go_backwards=False, initial_state=0, return_full_state=False, name='')
+
     Layer factory function to create a function that runs a step function recurrently over an input sequence.
     This implements the typical recurrent model.
 

bindings/python/cntk/losses/__init__.py

@@ -72,7 +72,8 @@ def cosine_distance_with_negative_samples(x, y, shift, num_negative_samples, nam
                [[ 1. ,  0. ,  0.5]]], dtype=float32)
 
     Args:
-        x, y: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
+        x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
+        y: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
         shift: non-zero positive integer representing number of shift to generate a negative sample
         num_negative_samples: number of negative samples to generate, a non-zero positive integer 
         name (str, optional): the name of the Function instance in the network

bindings/python/cntk/metrics/__init__.py

@@ -129,7 +129,9 @@ def edit_distance_error(input_a, input_b, subPen=1, delPen=1, insPen=1, squashIn
     Args:
         input_a: first input sequence
         input_b: second input sequence
-        subPen, delPen, insPen: substitution, deletion and insertion penalties
+        subPen: substitution penalty
+        delPen: deletion penalty
+        insPen: insertion penalty
         squashInputs: whether to merge sequences of identical samples (in both input sequences). If true and tokensToIgnore contains label '-' then
                 given first input sequence as s1="a-ab-" and second as s2="-aa--abb" the edit distance will be computed against s1' = "aab" and s2' = "aab".
         tokensToIgnore: list of samples to ignore during edit distance evaluation (in both sequences)

bindings/python/cntk/ops/__init__.py

@@ -335,7 +335,10 @@ def roipooling(conv_feature_map, rois, roi_output_shape, name=''):
 
 from cntk.cntk_py import PoolingType_Max, PoolingType_Average
 MAX_POOLING = PoolingType_Max
+'''int: constant used to specify maximum pooling'''
+
 AVG_POOLING = PoolingType_Average
+'''int: constant used to specify average pooling'''
 
 @typemap
 def pooling(operand, pooling_type, pooling_window_shape, strides=(1,), auto_padding=[False],
@@ -364,8 +367,8 @@ def pooling(operand, pooling_type, pooling_window_shape, strides=(1,), auto_padd
         auto_padding (default [False,]): automatic padding flags for each input dimension.
         lower_pad (default (0,)): precise lower padding for each input dimension
         upper_pad (default (0,)): precise upper padding for each input dimension
-        ceil_out_dim (default false): ceiling while computing output size
-        include_pad(default false): include pad while average pooling
+        ceil_out_dim (default False): ceiling while computing output size
+        include_pad(default False): include pad while average pooling
         name (str, optional): the name of the Function instance in the network
     Returns:
         :class:`~cntk.ops.functions.Function`
@@ -381,6 +384,8 @@ def pooling(operand, pooling_type, pooling_window_shape, strides=(1,), auto_padd
 
 
 MAX_UNPOOLING = PoolingType_Max
+'''int: constant used to specify maximum unpooling'''
+
 @typemap
 def unpooling(operand, pooling_input, unpooling_type, unpooling_window_shape, strides=(1,), auto_padding=[False],
             lower_pad=(0,), upper_pad=(0,), name=''):

bindings/python/cntk/ops/functions.py

@@ -932,7 +932,7 @@ def replace_placeholder(self, substitution):
         Returns:
             :class:`Function`: itself
 
-        :raises ExceptionType: when the function has multiple placeholders.
+        :raises Exception: when the function has multiple placeholders.
         '''
         return super(Function, self).replace_placeholder(substitution)
 

bindings/python/cntk/ops/variables.py

@@ -289,7 +289,7 @@ class Constant(VariableMixin, TensorOpsMixin, cntk_py.Constant):
     A constant value. It can be a scalar, vector, matrix, or tensor
     of floating point numbers that cannot be modified.
 
-    A Constant is a :class:`~cntk.ops.Variable` and therefore inherits all its methods.
+    A Constant is a :class:`~cntk.ops.variables.Variable` and therefore inherits all its methods.
 
     Args:
        value (`np.ndarray` or `list` or `float` or `int`): Initial value.

bindings/python/cntk/train/training_session.py

@@ -212,7 +212,7 @@ def on_cross_validation_end(self, index, average_error, num_samples, num_minibat
 @typemap
 def minibatch_size_schedule(schedule, epoch_size=1):
     '''
-    Create a minibatch size schedule
+    Creates a minibatch size schedule.
 
     Examples:
         >>> # Use a fixed value 32 for all minibatches
@@ -232,7 +232,7 @@ def minibatch_size_schedule(schedule, epoch_size=1):
         (32, 32, 64, 64, 128, 128)
 
     Args:
-        schedule (integer or list): if integer, it this minibatch size will be used for the whole training.
+        schedule (int or list): if integer, this minibatch size will be used for the whole training.
          In case of list of integers, the elements are used as the values for ``epoch_size`` samples. 
          If list contains pair, the second element is used as a value for (``epoch_size`` x first element) samples
         epoch_size (int): number of samples as a scheduling unit.

bindings/python/cntk/utils/__init__.py

@@ -143,7 +143,7 @@ def get_train_eval_criterion(trainer):
     Fetch the train evaluation criterion (e.g., classification error) from the last minibatch and copy it to the CPU in case it is on the GPU.
 
     Args:
-        trainer (:class:`Trainer`): the trainer used.
+        trainer (:class:`~cntk.train.trainer.Trainer`): the trainer used.
     Returns:
         the criterion value
     '''
@@ -201,7 +201,7 @@ def eval(op, arguments=None, precision=None, device=None, backward_pass=False, e
     mainly to explore the operators and for convenient unit testing.
 
     Args:
-        op (:class:`Function`): operation to evaluate
+        op (:class:`~cntk.ops.functions.Function`): operation to evaluate
         arguments: maps variables to their input data. The
          interpretation depends on the input type: