microsoft
/
CNTK
зеркало из
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность

Addressing CR comments

This commit is contained in:
Willi Richert 2016-11-11 09:17:50 +01:00
Родитель f8b5217dfc
Коммит b6d69c5866
9 изменённых файлов: 342 добавлений и 336 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

8
bindings/python/cntk/axis.py
Просмотреть файл

@ -41,7 +41,7 @@ class Axis(cntk_py.Axis):
Returns True if the axis is of type static and False otherwise
Returns:
`bool`: True if this axis is of type static and False otherwise
bool: True if this axis is of type static and False otherwise
'''
return super(Axis, self).is_static_axis()
@ -51,7 +51,7 @@ class Axis(cntk_py.Axis):
Returns the name of this axis.
Returns:
`str`: the name of this axis.
str: the name of this axis.
'''
return super(Axis, self).name()
@ -60,7 +60,7 @@ class Axis(cntk_py.Axis):
Returns the integer with which the static axis is defined. For example, 0 = first axis, 1 = second axis, etc.
Args:
checked (`bool`): if True then this function will throw an exception if the axis is not static.
checked (bool): if True then this function will throw an exception if the axis is not static.
Returns:
`int`: the number with which the static axis is defined.
@ -107,7 +107,7 @@ class Axis(cntk_py.Axis):
Creates an Axis object representing a new unique dynamic axis.
Args:
name (`str`): name of the dynmic axis
name (str): name of the dynmic axis
Returns:
:class:`Axis`: new unique dynamic axis

14
bindings/python/cntk/device.py
Просмотреть файл

@ -35,7 +35,7 @@ def all_devices():
Returns a device descriptor list with all the available devices
Returns:
:class:`cntk.device.DeviceDescriptor` list: all device descriptors
:class:`~cntk.device.DeviceDescriptor` list: all device descriptors
'''
return cntk_py.DeviceDescriptor.all_devices()
@ -44,7 +44,7 @@ def best():
Returns a device descriptor with the best configuration.
Returns:
:class:`cntk.device.DeviceDescriptor`: Best device descriptor
:class:`~cntk.device.DeviceDescriptor`: Best device descriptor
'''
return cntk_py.DeviceDescriptor.best_device()
@ -53,7 +53,7 @@ def cpu():
Returns CPU device descriptor
Returns:
:class:`cntk.device.DeviceDescriptor`: CPU device descriptor
:class:`~cntk.device.DeviceDescriptor`: CPU device descriptor
'''
return cntk_py.DeviceDescriptor.cpu_device()
@ -62,7 +62,7 @@ def default():
Returns default device
Returns:
:class:`cntk.device.DeviceDescriptor`: Default device descriptor
:class:`~cntk.device.DeviceDescriptor`: Default device descriptor
'''
return cntk_py.DeviceDescriptor.default_device()
@ -71,7 +71,7 @@ def gpu(device_id):
Returns GPU device
Returns:
:class:`cntk.device.DeviceDescriptor`: GPU device descriptor
:class:`~cntk.device.DeviceDescriptor`: GPU device descriptor
'''
return cntk_py.DeviceDescriptor.gpu_device(device_id)
@ -89,9 +89,9 @@ def set_default_device(new_default_device):
Set new device descriptor as default
Args:
new_default_device (:class:`cntk.device.DeviceDescriptor`): new device descriptor
new_default_device (:class:`~cntk.device.DeviceDescriptor`): new device descriptor
Returns:
:class:`cntk.device.DeviceDescriptor`: id
:class:`~cntk.device.DeviceDescriptor`: id
'''
return cntk_py.DeviceDescriptor.set_default_device(new_default_device)

80
bindings/python/cntk/io/__init__.py
Просмотреть файл

@ -57,9 +57,9 @@ class MinibatchSource(cntk_py.MinibatchSource):
Parent class of all minibatch sources. For most cases you will need the
helper functions :func:`text_format_minibatch_source` or
:func:`minibatch_source`.
A `MinibatchSource` can be indexed by a `StreamInfo`, which will return a
`MinibatchData` object that can be passed e.g. to the
:func:`cntk.trainer.Trainer.train_minibatch` function.
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.
'''
def __init__(self, deserializers=None, randomize=True, epoch_size=INFINITELY_REPEAT, distributed_communicator=None):
@ -81,7 +81,7 @@ class MinibatchSource(cntk_py.MinibatchSource):
Describes the stream that this source produces.
Returns:
`dict` mapping input names to the stream information
dict mapping input names to the stream information
'''
return super(MinibatchSource, self).stream_infos()
@ -98,7 +98,7 @@ class MinibatchSource(cntk_py.MinibatchSource):
Return the :class:`StreamInfo` for the given stream name
Args:
name (`str`): stream name to fetch :class:`StreamInfo` for
name (str): stream name to fetch :class:`StreamInfo` for
'''
return self.stream_info(name)
@ -115,11 +115,11 @@ class MinibatchSource(cntk_py.MinibatchSource):
when the MinibatchSource has no more data to return.
Args:
minibatch_size_in_samples (`int`): number of samples to retrieve for
minibatch_size_in_samples (int): number of samples to retrieve for
the next minibatch. Must be > 0.
minibatch_size_in_sequences (`int`, defaults to `None`): number of
minibatch_size_in_sequences (int, defaults to `None`): number of
samples to retrieve for the next minibatch. Must be > 0.
input_map (`dict`): mapping of :class:`~cntk.ops.variabls.Variable`
input_map (dict): mapping of :class:`~cntk.ops.variabls.Variable`
to :class:`StreamInformation` which will be used to convert the
returned data.
device (`DeviceDescriptor`, defaults to `None`): CNTK DeviceDescriptor
@ -179,7 +179,7 @@ def _py_dict_to_cntk_dict(py_dict):
'''
Converts a Python dictionary into a CNTK Dictionary whose values are CNTK DictionaryValue instances.
Args:
py_dict (`dict`): a dictionary to be converted.
py_dict (dict): a dictionary to be converted.
Returns:
:class:`~cntk_py.Dictionary`
'''
@ -208,7 +208,7 @@ def minibatch_source(config, distributed_communicator):
'''
Instantiate the CNTK built-in composite minibatch source which is used to stream data into the network.
Args:
config (`dict`): a dictionary containing all the key-value configuration entries.
config (dict): a dictionary containing all the key-value configuration entries.
distributed_communicator: optional distributed communicator
Returns:
:class:`MinibatchSource`
@ -227,8 +227,8 @@ class ReaderConfig(dict):
Args:
deserializers ('list', default is empty): list of deserializers
(:class:`ImageDeserializer` for now).
randomize (`bool`, default True): randomize images before every epoch
epoch_size (`int`): epoch size
randomize (bool, default True): randomize images before every epoch
epoch_size (int): epoch size
'''
def __init__(self, deserializers=None, randomize=True, epoch_size=INFINITELY_REPEAT):
@ -270,7 +270,7 @@ class Deserializer(dict):
========================== ============
Args:
type (`str`): type of the deserializer
type (str): type of the deserializer
See also:
https://github.com/microsoft/cntk/wiki/Understanding-and-Extending-Readers
@ -288,7 +288,7 @@ class ImageDeserializer(Deserializer):
<full path to image><tab><numerical label (0-based class id)>
Args:
filename (`str`): file name of the map file that associates images to
filename (str): file name of the map file that associates images to
classes
See also:
@ -322,7 +322,7 @@ class ImageDeserializer(Deserializer):
of the network with data augmentation.
Args:
node (`str` or input node): node or its name
node (str or input node): node or its name
transforms (`list` of transforms): the transforms can be created by
the static methods `crop`, `scale`, or `mean`.
@ -341,8 +341,8 @@ class ImageDeserializer(Deserializer):
ground truth of train or test.
Args:
node (`str` or input node): node or its name
num_classes (`int`): number of classes
node (str or input node): node or its name
num_classes (int): number of classes
'''
if not isinstance(node, str):
@ -355,7 +355,7 @@ class ImageDeserializer(Deserializer):
Crop transform that can be used to pass to `map_features`
Args:
crop_type (`str`, default 'center'): 'center' or 'random'. 'random'
crop_type (str, default 'center'): 'center' or 'random'. 'random'
is usually used during training while 'center' is usually for testing.
Random cropping is a popular data augmentation technique used to improve
generalization of the DNN.
@ -368,12 +368,12 @@ class ImageDeserializer(Deserializer):
augmentation technique), use colon-delimited values like cropRatio=0.875:0.466
which means 224 crop will be taken from images randomly scaled to have
size in [256, 480] range.
jitter_type (`str`, default 'uniRatio'): crop scale jitter type, possible
jitter_type (str, default 'uniRatio'): crop scale jitter type, possible
values are 'None', 'UniRatio'. 'uniRatio' means uniform distributed jitter
scale between the minimum and maximum cropRatio values.
Returns:
`dict` describing the crop transform
dict describing the crop transform
'''
return dict(type='Crop', cropType=crop_type, cropRatio=ratio,
jitterType=jitter_type)
@ -384,14 +384,14 @@ class ImageDeserializer(Deserializer):
Scale transform that can be used to pass to `map_features` for data augmentation.
Args:
width (`int`): width of the image in pixels
height (`int`): height of the image in pixels
channels (`int`): channels of the image
interpolations (`str`, default 'linear'): possible values are
width (int): width of the image in pixels
height (int): height of the image in pixels
channels (int): channels of the image
interpolations (str, default 'linear'): possible values are
'nearest', 'linear', 'cubic', and 'lanczos'
Returns:
`dict` describing the scale transform
dict describing the scale transform
'''
return dict(type='Scale', width=width, height=height, channels=channels,
interpolations=interpolations)
@ -402,11 +402,11 @@ class ImageDeserializer(Deserializer):
Mean transform that can be used to pass to `map_features` for data augmentation.
Args:
filename (`str`): file that stores the mean values for each pixel
filename (str): file that stores the mean values for each pixel
in OpenCV matrix XML format
Returns:
`dict` describing the mean transform
dict describing the mean transform
'''
return dict(type='Mean', meanFile=filename)
@ -427,7 +427,7 @@ class CTFDeserializer(Deserializer):
where
Sample=|Input_Name (Value )*
Args:
filename (`str`): file name containing the text input
filename (str): file name containing the text input
See also:
https://github.com/Microsoft/CNTK/wiki/CNTKTextFormat-Reader
'''
@ -451,12 +451,12 @@ class CTFDeserializer(Deserializer):
Example: for node name 'Apples' an input line could look like this:
|Apples 0 1 2 3 4 5 6 7 8 9
Args:
node (`str` or input node): node or its name
dim (`int`): specifies the dimension of the input value vector
node (str or input node): node or its name
dim (int): specifies the dimension of the input value vector
(for dense input this directly corresponds to the number of values in each sample,
for sparse this represents the upper bound on the range of possible index values).
format (`str`, default 'dense'): 'dense' or 'sparse'. Specifies the input type.
alias (`str`, default None): None or alias name. Optional abbreviated name that
format (str, default 'dense'): 'dense' or 'sparse'. Specifies the input type.
alias (str, default None): None or alias name. Optional abbreviated name that
is used in the text file to avoid repeating long input names. For details please
see https://github.com/Microsoft/CNTK/wiki/CNTKTextFormat-Reader
'''
@ -474,13 +474,13 @@ def text_format_minibatch_source(path, stream_configs, epoch_size=INFINITELY_REP
Creates a minibatch source from a CNTKTextFormatReader file.
Args:
path ('file'): filename of the data file
path (file): filename of the data file
stream_configs (`list` of :class:`StreamConfiguration` instances): list
of stream configurations, each of which describes one stream in the
file
epoch_size (`int`, optional): size of an epoch. In case of 0 the size
epoch_size (int, optional): size of an epoch. In case of 0 the size
of the training set will be taken. Default is max of 64bit.
randomize (`bool`, optional): whether to randomize the contents of data file.
randomize (bool, optional): whether to randomize the contents of data file.
distributed_communicator (:class:`~cntk.distributed.communicator`): optional distributed communicator
Returns:
@ -499,13 +499,13 @@ class StreamConfiguration(cntk_py.StreamConfiguration):
:func:`text_format_minibatch_source`.
Args:
name (`str`): name of this stream
dim (`int`): dimensions of this stream. A text format reader reads data
name (str): name of this stream
dim (int): dimensions of this stream. A text format reader reads data
as flat arrays. If you need different shapes you can
:func:`cntk.ops.reshape` it later.
is_sparse (`bool`, default `False`): whether the provided data is sparse
:func:`~cntk.ops.reshape` it later.
is_sparse (bool, default `False`): whether the provided data is sparse
(`False` by default)
stream_alias (`str`, default ''): name of the stream in the file that is fed to the
stream_alias (str, default ''): name of the stream in the file that is fed to the
:func:`text_format_minibatch_source`
'''

154
bindings/python/cntk/learner.py
Просмотреть файл

@ -66,7 +66,7 @@ class Learner(cntk_py.Learner):
gradient_values (`dict`): maps :class:`~cntk.variables.Parameter` to
a NumPy array containing the first order gradient values for the
Parameter w.r.t. the training objective.
training_sample_count (`int`): training sample count
training_sample_count (int): training sample count
Returns:
`False` to indicate that learning has stopped for all of the parameters associated with this learner
@ -91,7 +91,7 @@ class Learner(cntk_py.Learner):
Resets the learning rate.
Args:
learning_rate (`float`, `list` or a training schedule): learning rate
learning_rate (float, list or a training schedule): learning rate
to reset to
'''
learning_rate = learning_rate_schedule(learning_rate)
@ -102,7 +102,7 @@ class Learner(cntk_py.Learner):
The learning rate.
Args:
minibatch_size (``int``): minibatch size to re-scaled
minibatch_size (int): minibatch size to re-scaled
the learning rate to the per-sample value (in case when the schedule
was build with ``unit=UnitType.minibatch``).
'''
@ -132,11 +132,11 @@ def training_parameter_schedule(schedule, epoch_size=1, unit=UnitType.sample):
(0.1, 0.1, 0.01, 0.01, 0.001, 0.001)
Args:
schedule (``float`` or ``list``): if ``float``, is the parameter schedule to be used
schedule (float or list): if float, is the parameter schedule to be used
for all samples. In case of list, 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. Parameters in
epoch_size (int): number of samples as a scheduling unit. Parameters in
the schedule change their values every ``epoch_size`` samples.
unit (:class:`UnitType`): one of two
@ -145,6 +145,9 @@ def training_parameter_schedule(schedule, epoch_size=1, unit=UnitType.sample):
Returns:
training parameter schedule
See also:
:func:`learning_rate_schedule`
'''
if not isinstance(unit, UnitType):
raise ValueError('schedule unit "%s" is not supported' %
@ -176,15 +179,18 @@ def learning_rate_schedule(lr, epoch_size=1, unit=UnitType.sample):
:func:`training_parameter_schedule`).
Args:
lr (``float`` or ``list``): see parameter ``schedule`` in
lr (float or list): see parameter ``schedule`` in
:func:`training_parameter_schedule`.
epoch_size (``int``): see parameter ``epoch_size`` in
epoch_size (int): see parameter ``epoch_size`` in
:func:`training_parameter_schedule`.
unit (:class:`UnitType`): see parameter
``unit`` in :func:`training_parameter_schedule`.
Returns:
learning rate schedule
See also:
:func:`training_parameter_schedule`
'''
return training_parameter_schedule(lr, epoch_size, unit)
@ -195,9 +201,9 @@ def momentum_schedule(momentum, epoch_size=1, unit=UnitType.sample):
:func:`training_parameter_schedule`).
Args:
momentum (``float`` or ``list``): see parameter ``schedule`` in
momentum (float or list): see parameter ``schedule`` in
:func:`training_parameter_schedule`.
epoch_size (``int``): see parameter ``epoch_size`` in
epoch_size (int): see parameter ``epoch_size`` in
:func:`training_parameter_schedule`.
unit (:class:`UnitType`): see parameter
``unit`` in :func:`training_parameter_schedule`.
@ -223,9 +229,9 @@ def momentum_schedule(momentum, epoch_size=1, unit=UnitType.sample):
(0.99, 0.99, 0.88, 0.88, 0.77)
Args:
momentum (``float`` or ``list``): see parameter ``schedule`` in
momentum (float or list): see parameter ``schedule`` in
:func:`training_parameter_schedule`.
epoch_size (``int``): see parameter ``epoch_size`` in
epoch_size (int): see parameter ``epoch_size`` in
:func:`training_parameter_schedule`.
unit (:class:`UnitType`): see parameter
``unit`` in :func:`training_parameter_schedule`.
@ -242,9 +248,9 @@ def momentum_as_time_constant_schedule(momentum, epoch_size=1):
semantics as :func:`training_parameter_schedule`).
Args:
momentum (``float`` or ``list``): see parameter ``schedule`` in
momentum (float or list): see parameter ``schedule`` in
:func:`training_parameter_schedule`.
epoch_size (``int``): see parameter ``epoch_size`` in
epoch_size (int): see parameter ``epoch_size`` in
:func:`training_parameter_schedule`.
unit (:class:`UnitType`): see parameter
``unit`` in :func:`training_parameter_schedule`.
@ -267,9 +273,9 @@ def momentum_as_time_constant_schedule(momentum, epoch_size=1):
>>> m = momentum_as_time_constant_schedule([1100, 1500], 1000)
Args:
momentum (``float`` or ``list``): see parameter ``schedule`` in
momentum (float or list): see parameter ``schedule`` in
:func:`training_parameter_schedule`.
epoch_size (``int``): see parameter ``epoch_size`` in
epoch_size (int): see parameter ``epoch_size`` in
:func:`training_parameter_schedule`.
Returns:
@ -299,21 +305,21 @@ def sgd(parameters, lr,
information on how to set the parameters.
Args:
parameters (`list` of parameters): list of network parameters to tune.
parameters (list of parameters): list of network parameters to tune.
These can be obtained by the '.parameters()' method of the root
operator.
lr (``float``, ``list`` or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a `float` or a `list`, lr is
lr (float, list or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a float or a list, lr is
converted to a per-sample schedule by invoking :func:`learning_rate_schedule`.
l1_regularization_weight (``float``, optional): the L1 regularization weight per sample,
l1_regularization_weight (float, optional): the L1 regularization weight per sample,
defaults to 0.0
l2_regularization_weight (``float``, optional): the L2 regularization weight per sample,
l2_regularization_weight (float, optional): the L2 regularization weight per sample,
defaults to 0.0
gaussian_noise_injection_std_dev (``float``, optional): the standard deviation
gaussian_noise_injection_std_dev (float, optional): the standard deviation
of the Gaussian noise added to parameters post update, defaults to 0.0
gradient_clipping_threshold_per_sample (``float``, optional): clipping threshold
gradient_clipping_threshold_per_sample (float, optional): clipping threshold
per sample, defaults to infinity
gradient_clipping_with_truncation (``bool``, default ``True``): gradient clipping
gradient_clipping_with_truncation (bool, default ``True``): gradient clipping
Returns:
Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
@ -346,22 +352,22 @@ def momentum_sgd(parameters, lr, momentum,
Args:
parameters (list of parameters): list of network parameters to tune.
These can be obtained by the root operator's ``parameters``.
lr (``float``, `list````` or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a `float` or a `list`, lr is
lr (float, list```` or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a float or a list, lr is
converted to a per-sample schedule by invoking :func:`learning_rate_schedule`.
momentum (``float``, ``list`` or output of :func:`momentum_schedule` or :func:`momentum_as_time_constant_schedule`): momentum schedule. When the argument
value is a ``float`` or a ``list``, momentum is converted to a per-sample schedule by
momentum (float, list or output of :func:`momentum_schedule` or :func:`momentum_as_time_constant_schedule`): momentum schedule. When the argument
value is a float or a list, momentum is converted to a per-sample schedule by
invoking :func:`momentum_schedule`. Refer to the `wiki
<https://github.com/Microsoft/CNTK/wiki/SGD-block#converting-learning-rate-and-momentum-parameters-from-other-toolkits>`_.
l1_regularization_weight (``float``, optional): the L1 regularization weight per sample,
l1_regularization_weight (float, optional): the L1 regularization weight per sample,
defaults to 0.0
l2_regularization_weight (``float``, optional): the L2 regularization weight per sample,
l2_regularization_weight (float, optional): the L2 regularization weight per sample,
defaults to 0.0
gaussian_noise_injection_std_dev (``float``, optional): the standard deviation
gaussian_noise_injection_std_dev (float, optional): the standard deviation
of the Gaussian noise added to parameters post update, defaults to 0.0
gradient_clipping_threshold_per_sample (``float``, optional): clipping threshold
gradient_clipping_threshold_per_sample (float, optional): clipping threshold
per sample, defaults to infinity
gradient_clipping_with_truncation (``bool``, default ``True``): gradient clipping
gradient_clipping_with_truncation (bool, default ``True``): gradient clipping
Returns:
Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
@ -387,28 +393,28 @@ def nesterov(parameters, lr, momentum,
gradient_clipping_with_truncation=True):
'''
Creates a Nesterov SGD learner instance to learn the parameters. This was
originally proposed by Nesterov [1] in 1983 and then proved to work well in
originally proposed by Nesterov [1] in 1983 and then shown to work well in
a deep learning context by Sutskever, et al. [2].
Args:
parameters (list of parameters): list of network parameters to tune.
These can be obtained by the root operator's ``parameters``.
lr (``float``, ``list`` or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a ``float`` or a ``list``, lr is
lr (float, list or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a float or a list, lr is
converted to a per-sample schedule by invoking :func:`learning_rate_schedule`.
momentum (``float``, ``list`` or output of :func:`momentum_schedule` or :func:`momentum_as_time_constant_schedule`): momentum schedule. When the argument
value is a ``float`` or a ``list``, momentum is converted to a per-sample schedule by
momentum (float, list or output of :func:`momentum_schedule` or :func:`momentum_as_time_constant_schedule`): momentum schedule. When the argument
value is a float or a list, momentum is converted to a per-sample schedule by
invoking :func:`momentum_schedule`. Refer to the `wiki
<https://github.com/Microsoft/CNTK/wiki/SGD-block#converting-learning-rate-and-momentum-parameters-from-other-toolkits>`_.
l1_regularization_weight (``float``, optional): the L1 regularization weight per sample,
l1_regularization_weight (float, optional): the L1 regularization weight per sample,
defaults to 0.0
l2_regularization_weight (``float``, optional): the L2 regularization weight per sample,
l2_regularization_weight (float, optional): the L2 regularization weight per sample,
defaults to 0.0
gaussian_noise_injection_std_dev (``float``, optional): the standard deviation
gaussian_noise_injection_std_dev (float, optional): the standard deviation
of the Gaussian noise added to parameters post update, defaults to 0.0
gradient_clipping_threshold_per_sample (``float``, optional): clipping threshold
gradient_clipping_threshold_per_sample (float, optional): clipping threshold
per sample, defaults to infinity
gradient_clipping_with_truncation (``bool``, default ``True``): gradient clipping
gradient_clipping_with_truncation (bool, default ``True``): gradient clipping
Returns:
Instance of a :class:`~cntk.learner.Learner` that can be passed to the
@ -449,19 +455,19 @@ def adagrad(parameters, lr, need_ave_multiplier=True,
Args:
parameters (list of parameters): list of network parameters to tune.
These can be obtained by the root operator's ``parameters``.
lr (``float``, `list` or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a `float` or a `list`, lr is
lr (float, list or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a float or a list, lr is
converted to a per-sample schedule by invoking :func:`learning_rate_schedule`.
need_ave_multiplier (``bool``, default):
l1_regularization_weight (``float``, optional): the L1 regularization weight per sample,
need_ave_multiplier (bool, default):
l1_regularization_weight (float, optional): the L1 regularization weight per sample,
defaults to 0.0
l2_regularization_weight (``float``, optional): the L2 regularization weight per sample,
l2_regularization_weight (float, optional): the L2 regularization weight per sample,
defaults to 0.0
gaussian_noise_injection_std_dev (``float``, optional): the standard deviation
gaussian_noise_injection_std_dev (float, optional): the standard deviation
of the Gaussian noise added to parameters post update, defaults to 0.0
gradient_clipping_threshold_per_sample (``float``, optional): clipping threshold
gradient_clipping_threshold_per_sample (float, optional): clipping threshold
per sample, defaults to infinity
gradient_clipping_with_truncation (``bool``, default `True`): gradient clipping
gradient_clipping_with_truncation (bool, default `True`): gradient clipping
Returns:
Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
@ -500,25 +506,25 @@ def adam_sgd(parameters, lr, momentum,
Args:
parameters (list of parameters): list of network parameters to tune.
These can be obtained by the root operator's ``parameters``.
lr (``float``, `list` or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a `float` or a `list`, lr is
lr (float, list or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a float or a list, lr is
converted to a per-sample schedule by invoking :func:`learning_rate_schedule`.
momentum (`float`, `list` or output of :func:`momentum_schedule` or :func:`momentum_as_time_constant_schedule`): momentum schedule. When the argument
value is a `float` or a `list`, momentum is converted to a per-sample schedule by
momentum (float, list or output of :func:`momentum_schedule` or :func:`momentum_as_time_constant_schedule`): momentum schedule. When the argument
value is a float or a list, momentum is converted to a per-sample schedule by
invoking :func:`momentum_schedule`. Refer to the `wiki
<https://github.com/Microsoft/CNTK/wiki/SGD-block#converting-learning-rate-and-momentum-parameters-from-other-toolkits>`_.
variance_momentum (`float`, `list` or output of :func:`momentum_schedule` or :func:`momentum_as_time_constant_schedule`): variance momentum schedule. When the argument
value is a `float` or a `list`, variance momentum is converted to a per-sample schedule by
variance_momentum (float, list or output of :func:`momentum_schedule` or :func:`momentum_as_time_constant_schedule`): variance momentum schedule. When the argument
value is a float or a list, variance momentum is converted to a per-sample schedule by
invoking :func:`momentum_schedule`. Defaults to momentum_as_time_constant_schedule(720000).
l1_regularization_weight (``float``, optional): the L1 regularization weight per sample,
l1_regularization_weight (float, optional): the L1 regularization weight per sample,
defaults to 0.0
l2_regularization_weight (``float``, optional): the L2 regularization weight per sample,
l2_regularization_weight (float, optional): the L2 regularization weight per sample,
defaults to 0.0
gaussian_noise_injection_std_dev (``float``, optional): the standard deviation
gaussian_noise_injection_std_dev (float, optional): the standard deviation
of the Gaussian noise added to parameters post update, defaults to 0.0
gradient_clipping_threshold_per_sample (``float``, optional): clipping threshold
gradient_clipping_threshold_per_sample (float, optional): clipping threshold
per sample, defaults to infinity
gradient_clipping_with_truncation (``bool``, default `True`): gradient clipping
gradient_clipping_with_truncation (bool, default `True`): gradient clipping
Returns:
Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`
@ -559,24 +565,24 @@ def rmsprop(parameters, lr,
Args:
parameters (list of parameters): list of network parameters to tune.
These can be obtained by the root operator's ``parameters``.
lr (``float``, `list` or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a `float` or a `list`, lr is
lr (float, list or output of :func:`learning_rate_schedule`): learning rate
schedule. When the argument value is a float or a list, lr is
converted to a per-sample schedule by invoking :func:`learning_rate_schedule`.
gamma (``float``):
inc (``float``):
dec (``float``):
max (``float``):
min (``float``):
need_ave_multiplier (``bool``, default):
l1_regularization_weight (``float``, optional): the L1 regularization weight per sample,
gamma (float):
inc (float):
dec (float):
max (float):
min (float):
need_ave_multiplier (bool, default):
l1_regularization_weight (float, optional): the L1 regularization weight per sample,
defaults to 0.0
l2_regularization_weight (``float``, optional): the L2 regularization weight per sample,
l2_regularization_weight (float, optional): the L2 regularization weight per sample,
defaults to 0.0
gaussian_noise_injection_std_dev (``float``, optional): the standard deviation
gaussian_noise_injection_std_dev (float, optional): the standard deviation
of the Gaussian noise added to parameters post update, defaults to 0.0
gradient_clipping_threshold_per_sample (``float``, optional): clipping threshold
gradient_clipping_threshold_per_sample (float, optional): clipping threshold
per sample, defaults to infinity
gradient_clipping_with_truncation (``bool``, default `True`): gradient clipping
gradient_clipping_with_truncation (bool, default `True`): gradient clipping
Returns:
Instance of a :class:`~cntk.learner.Learner` that can be passed to the :class:`~cntk.trainer.Trainer`

340
bindings/python/cntk/ops/__init__.py
Просмотреть файл

@ -20,11 +20,11 @@ def combine(operands, name=''):
with 2 outputs; viz. CrossEntropy loss and ClassificationError output.
Args:
operands (`list`): list of functions or their variables to combine
name (`str`, optional): the name of the Combine Function in the network
operands (list): list of functions or their variables to combine
name (str, optional): the name of the Combine Function in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import combine
converted_operands = list()
@ -49,10 +49,10 @@ def alias(x, name=''):
Args:
operand: The Function/Variable to alias
name (`str`, optional): the name of the Alias Function in the network
name (str, optional): the name of the Alias Function in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import alias
x = sanitize_input(x)
@ -73,9 +73,9 @@ def binary_cross_entropy(output, target, name=''):
Args:
output: the computed posterior probability from the network
target: ground-truth label, 0 or 1
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import binary_cross_entropy
dtype = get_data_type(output, target)
@ -95,9 +95,9 @@ def weighted_binary_cross_entropy(output, target, weight, name=''):
output: the computed posterior probability from the network
target: ground-truth label, 0 or 1
weight: weight of each example
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import weighted_binary_cross_entropy
dtype = get_data_type(output, target, weight)
@ -134,11 +134,11 @@ def cross_entropy_with_softmax(output_vector, target_vector, axis=-1, name=''):
target_vector: usually it is one-hot vector where the hot bit
corresponds to the label index. But it can be any probability
distribution over the labels.
axis (`int` or :class:`cntk.axis.Axis`): axis along which the cross
axis (int or :class:`~cntk.axis.Axis`): axis along which the cross
entropy will be computed.
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import cross_entropy_with_softmax
dtype = get_data_type(output_vector, target_vector)
@ -168,9 +168,9 @@ def squared_error(output, target, name=''):
output: the output values from the network
target: it is usually a one-hot vector where the hot bit
corresponds to the label index
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import squared_error
dtype = get_data_type(output, target)
@ -204,11 +204,11 @@ def classification_error(output_vector, target_vector, axis=-1, topN=1, name='')
output_vector: the output values from the network
target_vector: it is one-hot vector where the hot bit corresponds to
the label index.
axis (`int` or :class:`cntk.axis.Axis`): axis along which the
axis (int or :class:`~cntk.axis.Axis`): axis along which the
classification error will be computed.
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import classification_error
dtype = get_data_type(output_vector, target_vector)
@ -260,7 +260,7 @@ def convolution(convolution_map, operand, strides=(1,), sharing=[True],
convolution_map: convolution filter weights, stored as a tensor of dimensions :math:`[O \\times I \\times m_1 \\times m_2 \\times \\ldots \\times m_n]`,
where :math:`[m_1 \\times m_2 \\times \\ldots \\times m_n]` must be the kernel dimensions (spatial extent of the filter).
operand: convolution input. A tensor with dimensions :math:`[I \\times M_1 \\times M_2 \\times \\ldots \\times M_n]`.
strides (`tuple`, optional): stride dimensions. If strides[i] > 1 then only pixel positions that are multiples of strides[i] are computed.
strides (tuple, optional): stride dimensions. If strides[i] > 1 then only pixel positions that are multiples of strides[i] are computed.
For example, a stride of 2 will lead to a halving of that dimension. The first stride dimension that lines up with the number
of input channels can be set to any non-zero value.
sharing (bool): sharing flags for each input dimension
@ -276,9 +276,9 @@ def convolution(convolution_map, operand, strides=(1,), sharing=[True],
operations. Some convolution engines (e.g. cuDNN and GEMM-based engines) can benefit from using workspace as it may improve
performance. However, sometimes this may lead to higher memory utilization. Default is 0 which means the same as the input
samples.
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import convolution
operand = sanitize_input(operand)
@ -301,9 +301,9 @@ def roipooling(conv_feature_map, rois, roi_output_shape, name=''):
conv_feature_map: a convolutional feature map as the input volume ([W x H x C x N]).
rois: the coordinates of the ROIs per image ([4 x roisPerImage x N]), each ROI is (x, y, w, h) relative to original image size.
roi_output_shape: dimensions (width x height) of the ROI pooling output shape
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import roipooling
conv_feature_map = sanitize_input(conv_feature_map)
@ -344,9 +344,9 @@ def pooling(operand, pooling_type, pooling_window_shape, strides=(1,), auto_padd
auto_padding: automatic padding flags for each input dimension.
lower_pad: precise lower padding for each input dimension
upper_pad: precise upper padding for each input dimension
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import pooling
operand = sanitize_input(operand)
@ -376,17 +376,17 @@ def batch_normalization(operand, scale, bias, running_mean, running_inv_std, spa
training as well. You must pass a parameter tensor with initial value 0 and the same dimensions
as ``scale`` and ``bias``
running_inv_std: running variance. Represented as ``running_mean``
spatial(`bool`): flag that indicates whether to compute mean/var for each feature in a minibatch
spatial(bool): flag that indicates whether to compute mean/var for each feature in a minibatch
independently or, in case of convolutional layers, per future map
normalization_time_constant(`float`, default 5000): time constant for computing running average of
normalization_time_constant(float, default 5000): time constant for computing running average of
mean and variance as a low-pass filtered version of the batch statistics.
blend_time_constant(`float`, default 0): constant for smoothing batch estimates with the running
blend_time_constant(float, default 0): constant for smoothing batch estimates with the running
statistics
epsilon: conditioner constant added to the variance when computing the inverse standard deviation
use_cudnn_engine(`bool`, default True):
name (`str`, optional): the name of the Function instance in the network
use_cudnn_engine(bool, default True):
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import batch_normalization
operand = sanitize_input(operand)
@ -414,9 +414,9 @@ def less(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import less
dtype = get_data_type(left, right)
@ -440,9 +440,9 @@ def equal(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import equal
dtype = get_data_type(left, right)
@ -466,9 +466,9 @@ def greater(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import greater
dtype = get_data_type(left, right)
@ -492,9 +492,9 @@ def greater_equal(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import greater_equal
dtype = get_data_type(left, right)
@ -518,9 +518,9 @@ def not_equal(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import not_equal
dtype = get_data_type(left, right)
@ -544,9 +544,9 @@ def less_equal(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import less_equal
dtype = get_data_type(left, right)
@ -574,9 +574,9 @@ def plus(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import plus
dtype = get_data_type(left, right)
@ -601,9 +601,9 @@ def minus(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import minus
@ -629,9 +629,9 @@ def element_times(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import element_times
dtype = get_data_type(left, right)
@ -656,9 +656,9 @@ def element_divide(left, right, name=''):
Args:
left: left side tensor
right: right side tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import element_divide
dtype = get_data_type(left, right)
@ -700,14 +700,14 @@ def times(left, right, output_rank=1, infer_input_rank_to_map=-1, name=''):
Args:
left: left side matrix or tensor
right: right side matrix or tensor
output_rank (`int`): in case we have tensors as arguemnts, output_rank represents
output_rank (int): in case we have tensors as arguemnts, output_rank represents
the number of axes to be collapsed in order to transform the tensors
into matrices, perform the operation and then reshape back (explode the axes)
infer_input_rank_to_map ('int'): meant for internal use only. Always use default value
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import times
dtype = get_data_type(left, right)
@ -801,10 +801,10 @@ def times_transpose(left, right, name=''):
Args:
left: left side tensor
right: right side matrix or vector
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import times_transpose
dtype = get_data_type(left, right)
@ -841,9 +841,9 @@ def floor(arg, name=''):
Args:
arg: input tensor
name (`str`, optional): the name of the Function instance in the network (optional)
name (str, optional): the name of the Function instance in the network (optional)
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import floor
arg = sanitize_input(arg, get_data_type(arg))
@ -866,9 +866,9 @@ def ceil(arg, name=''):
Args:
arg: input tensor
name (`str`, optional): the name of the Function instance in the network (optional)
name (str, optional): the name of the Function instance in the network (optional)
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import ceil
arg = sanitize_input(arg, get_data_type(arg))
@ -901,9 +901,9 @@ def round(arg, name=''):
Args:
arg: input tensor
name (`str`, optional): the name of the Function instance in the network (optional)
name (str, optional): the name of the Function instance in the network (optional)
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import round
arg = sanitize_input(arg, get_data_type(arg))
@ -931,13 +931,13 @@ def clip(x, min_value, max_value, name=''):
Args:
x: tensor to be clipped
min_value (`float`): a scalar or a tensor which represents the minimum value to clip element
min_value (float): a scalar or a tensor which represents the minimum value to clip element
values to
max_value (`float`): a scalar or a tensor which represents the maximum value to clip element
max_value (float): a scalar or a tensor which represents the maximum value to clip element
values to
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import clip
x = sanitize_input(x, get_data_type(x))
@ -959,10 +959,10 @@ def relu(x, name=''):
array([[ 0., 0., 0., 1., 2.]], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import re_lu
x = sanitize_input(x)
@ -983,10 +983,10 @@ def sigmoid(x, name=''):
array([ 0.119203, 0.268941, 0.5 , 0.731059, 0.880797], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import sigmoid
x = sanitize_input(x)
@ -1006,10 +1006,10 @@ def tanh(x, name=''):
[ 0.995055, 0.999329]], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import tanh
x = sanitize_input(x)
@ -1038,10 +1038,10 @@ def softmax(x, name=''):
array([ 0.5, 0.5], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import softmax
x = sanitize_input(x)
@ -1062,10 +1062,10 @@ def hardmax(x, name=''):
array([ 0., 1., 0., 0.], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`): the name of the node in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str): the name of the node in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import hardmax
x = sanitize_input(x)
@ -1084,10 +1084,10 @@ def exp(x, name=''):
array([ 1. , 2.718282], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import exp
x = sanitize_input(x)
@ -1104,10 +1104,10 @@ def log(x, name=''):
array([ 0. , 0.693147], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
Note:
CNTK returns -85.1 for log(x) if ``x`` is negative or zero. The reason is that
@ -1132,10 +1132,10 @@ def sqrt(x, name=''):
array([ 0., 2.], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
Note:
CNTK returns zero for sqrt of negative nubmers, this will be changed to
@ -1156,10 +1156,10 @@ def square(x, name=''):
array([ 1., 100.], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import square
x = sanitize_input(x)
@ -1178,10 +1178,10 @@ def abs(x, name=''):
array([ 1., 1., 2., 3.], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import abs
x = sanitize_input(x)
@ -1200,10 +1200,10 @@ def negate(x, name=''):
array([ 1., -1., 2., -3.], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import negate
x = sanitize_input(x)
@ -1220,10 +1220,10 @@ def reciprocal(x, name=''):
array([-3. , 5. , -0.5 , 0.333333], dtype=float32)
Args:
x: numpy array or any :class:`cntk.ops.functions.Function` that outputs a tensor
name (`str`, optional): the name of the Function instance in the network
x: numpy array or any :class:`~cntk.ops.functions.Function` that outputs a tensor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import reciprocal
x = sanitize_input(x)
@ -1245,9 +1245,9 @@ def element_select(flag, value_if_true, value_if_false, name=''):
flag: condition tensor
value_if_true: true branch tensor
value_if_false: false branch tensor
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import element_select
flag = sanitize_input(flag)
@ -1295,10 +1295,10 @@ def future_value(x, initial_state=None, time_step=1, name=''):
x: the tensor (or its name) from which the future value is obtained.
initial_state: tensor or scalar representing the initial value to be
used when the input tensor is shifted in time.
time_step (`int`): the number of time steps to look into the future (default 1)
name (`str`, optional): the name of the Function instance in the network
time_step (int): the number of time steps to look into the future (default 1)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from ..utils import sanitize_dtype_cntk
@ -1347,11 +1347,11 @@ def past_value(x, initial_state=None, time_step=1, name=''):
x: the tensor (or its name) from which the past value is obtained
initial_state: tensor or scalar representing the initial value to be
used when the input tensor is shifted in time.
time_step (`int`): the number of time steps to look into the past (default 1)
name (`str`, optional): the name of the Function instance in the network
time_step (int): the number of time steps to look into the past (default 1)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from ..utils import sanitize_dtype_cntk
@ -1387,10 +1387,10 @@ def reshape(x, shape, name=''):
Args:
x: tensor to be reshaped
shape (`tuple`): a tuple defining the resulting shape
name (`str`, optional): the name of the Function instance in the network
shape (tuple): a tuple defining the resulting shape
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
if np.any(np.asarray(shape) < 0):
# TODO decide on whether -1 instead of 0 should be used to infer the
@ -1417,11 +1417,11 @@ def transpose(x, axis1=0, axis2=1, name=''):
Args:
x: tensor to be transposed
axis1 (`int` or :class:`cntk.axis.Axis`): the axis to swap with ``axis2``
axis2 (`int` or :class:`cntk.axis.Axis`): the axis to swap with ``axis1``
name (`str`, optional): the name of the Function instance in the network
axis1 (int or :class:`~cntk.axis.Axis`): the axis to swap with ``axis2``
axis2 (int or :class:`~cntk.axis.Axis`): the axis to swap with ``axis1``
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import transpose_axes
x = sanitize_input(x)
@ -1471,17 +1471,17 @@ def slice(x, axis, begin_index, end_index, name=''):
Args:
x: input tensor
axis (`int` or :class:`cntk.axis.Axis`): axis along which ``begin_index`` and ``end_index``
will be used. If it is of type `int` it will be used as a static axis.
begin_index (`int`): the index along axis where the slicing starts
end_index (`int`): the index along axis where the slicing ends
name (`str`, optional): the name of the Function instance in the network
axis (int or :class:`~cntk.axis.Axis`): axis along which ``begin_index`` and ``end_index``
will be used. If it is of type int it will be used as a static axis.
begin_index (int): the index along axis where the slicing starts
end_index (int): the index along axis where the slicing ends
name (str, optional): the name of the Function instance in the network
See also:
Indexing in NumPy: http://docs.scipy.org/doc/numpy/reference/arrays.indexing.html
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import slice
x = sanitize_input(x)
@ -1516,13 +1516,13 @@ def splice(inputs, axis=-1, name=''):
[ 50., 60.]]], dtype=float32)
Args:
inputs (`list`): tuple of input tensors
axis (`int` or :class:`cntk.axis.Axis`): axis along which the
inputs (list): tuple of input tensors
axis (int or :class:`~cntk.axis.Axis`): axis along which the
concatenation will be performed
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import splice
if type(inputs) not in (list, tuple):
@ -1576,11 +1576,11 @@ def reduce_sum(x, axis=None, name=''):
Args:
x: input tensor
axis (`int` or :class:`cntk.axis.Axis`): axis along which the reduction will be performed
name (`str`, optional): the name of the Function instance in the network
axis (int or :class:`~cntk.axis.Axis`): axis along which the reduction will be performed
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import reduce_sum
x = sanitize_input(x)
@ -1605,11 +1605,11 @@ def reduce_log_sum(x, axis=None, name=''):
Args:
x: input tensor
axis (`int` or :class:`cntk.axis.Axis`): axis along which the reduction will be performed
name (`str`): the name of the Function instance in the network
axis (int or :class:`~cntk.axis.Axis`): axis along which the reduction will be performed
name (str): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import reduce_log_sum
x = sanitize_input(x)
@ -1636,11 +1636,11 @@ def reduce_mean(x, axis=None, name=''):
Args:
x: input tensor
axis (`int` or :class:`cntk.axis.Axis`): axis along which the reduction will be performed
name (`str`, optional): the name of the Function instance in the network
axis (int or :class:`~cntk.axis.Axis`): axis along which the reduction will be performed
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import reduce_mean
x = sanitize_input(x)
@ -1667,11 +1667,11 @@ def reduce_max(x, axis=None, name=''):
Args:
x: input tensor
axis (`int` or :class:`cntk.axis.Axis`): axis along which the reduction will be performed
name (`str`): the name of the Function instance in the network
axis (int or :class:`~cntk.axis.Axis`): axis along which the reduction will be performed
name (str): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import reduce_max
x = sanitize_input(x)
@ -1698,11 +1698,11 @@ def reduce_min(x, axis=None, name=''):
Args:
x: input tensor
axis (`int` or :class:`cntk.axis.Axis`): axis along which the reduction will be performed
name (`str`): the name of the Function instance in the network
axis (int or :class:`~cntk.axis.Axis`): axis along which the reduction will be performed
name (str): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import reduce_min
x = sanitize_input(x)
@ -1732,12 +1732,12 @@ def random_sample(weights, num_samples, allow_duplicates, name=''):
Args:
weights: input vector of sampling weights which should be
non-negative numbers.
num_samples (`int`): number of expected samples
allow_duplicates (`bool`): If sampling is done
num_samples (int): number of expected samples
allow_duplicates (bool): If sampling is done
with replacement (`True`) or without (`False`).
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import random_sample
@ -1766,8 +1766,8 @@ def random_sample_inclusion_frequency(
Args:
weights: input vector of sampling weights which should be
non-negative numbers.
num_samples (`int`): number of expected samples
allow_duplicates (`bool`): If sampling is done
num_samples (int): number of expected samples
allow_duplicates (bool): If sampling is done
with replacement (`True`) or without (`False`).
Examples:
@ -1792,7 +1792,7 @@ def random_sample_inclusion_frequency(
1.0
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import random_sample_inclusion_frequency
weights = sanitize_input(weights)
@ -1830,11 +1830,11 @@ def dropout(x, dropout_rate=0.0, name=''):
Args:
x: input tensor
dropout_rate (`float`, [0,1)): probability that an element of ``x`` will be set to zero
name (:class:`str`, optional): the name of the Function instance in the network
dropout_rate (float, [0,1)): probability that an element of ``x`` will be set to zero
name (:class:str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
if dropout_rate < 0.0 or dropout_rate >= 1.0:
raise ValueError('dropout_rate must be in the interval [0,1)')
@ -1862,15 +1862,15 @@ def input_variable(shape, dtype=np.float32, needs_gradient=True, is_sparse=False
It creates an input node.
Args:
shape (`tuple` or `int`): the shape of the input tensor
dtype (`type`, optional): np.float32 (default) or np.float64
needs_gradients (`bool`, optional): whether to back-propagates to it or not. True by default.
is_sparse (`bool`, optional): whether the variable is sparse (`False` by default)
dynamic_axes (`list` or `tuple`, default): a list of dynamic axis (e.g., batch axis, time axis)
name (`str`, optional): the name of the Function instance in the network
shape (tuple or int): the shape of the input tensor
dtype (type, optional): np.float32 (default) or np.float64
needs_gradients (bool, optional): whether to back-propagates to it or not. True by default.
is_sparse (bool, optional): whether the variable is sparse (`False` by default)
dynamic_axes (list or tuple, default): a list of dynamic axis (e.g., batch axis, time axis)
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.variables.Variable`
:class:`~cntk.ops.variables.Variable`
'''
from cntk.cntk_py import input_variable
from ..utils import sanitize_shape, sanitize_dtype_cntk
@ -1895,11 +1895,11 @@ def placeholder_variable(shape=None, dynamic_axes=None, name=''):
are unfolded, the place holder will get assigned a variable along the correspondent dynamic axis.
Args:
shape (`tuple` or `int`): the shape of the variable tensor
dynamic_axes (`list`): the list of dynamic axes that the actual variable uses
shape (tuple or int): the shape of the variable tensor
dynamic_axes (list): the list of dynamic axes that the actual variable uses
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import placeholder_variable, NDShape, Axis
@ -1928,7 +1928,7 @@ def parameter(shape=None, init=None, device=None, name=''):
[ 2., 2., 2., 2.]], dtype=float32)
Args:
shape (`tuple` or `int`, optional): the shape of the input tensor. If not provided, it
shape (tuple or int, optional): the shape of the input tensor. If not provided, it
will be inferred from ``value``.
init (scalar or NumPy array or initializer): if init is a scalar
it will be replicated for every element in the tensor or
@ -1936,11 +1936,11 @@ def parameter(shape=None, init=None, device=None, name=''):
:mod:`cntk.initializer` it will be used to initialize the tensor at
the first forward pass. If `None`, the tensor will be initialized
with 0.
device (:class:`cntk.device.DeviceDescriptor`): instance of DeviceDescriptor
name (`str`, optional): the name of the Parameter instance in the network
device (:class:`~cntk.device.DeviceDescriptor`): instance of DeviceDescriptor
name (str, optional): the name of the Parameter instance in the network
Returns:
:class:`cntk.ops.variables.Parameter`
:class:`~cntk.ops.variables.Parameter`
'''
from .variables import Parameter
@ -1975,12 +1975,12 @@ def constant(value=None, shape=None, device=None, name=''):
value (scalar or NumPy array, optional): a scalar initial value that would be replicated for
every element in the tensor or NumPy array.
If ``None``, the tensor will be initialized uniformly random.
shape (`tuple` or `int`, optional): the shape of the input tensor. If not provided, it will
shape (tuple or int, optional): the shape of the input tensor. If not provided, it will
be inferred from ``value``.
device (:class:`cntk.device.DeviceDescriptor`): instance of DeviceDescriptor
name (`str`, optional): the name of the Function instance in the network
device (:class:`~cntk.device.DeviceDescriptor`): instance of DeviceDescriptor
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.variables.Constant`
:class:`~cntk.ops.variables.Constant`
'''
from .variables import Constant
if not device:
@ -2011,9 +2011,9 @@ def per_dim_mean_variance_normalize(operand, mean, inv_stddev, name=''):
operand: the variable to be normalized
mean (NumPy array): per dimension mean to use for the normalization
inv_stddev (NumPy array): per dimension standard deviation to use for the normalization
name (`str`, optional): the name of the Function instance in the network
name (str, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import per_dim_mean_variance_normalize
mean = sanitize_input(mean, get_data_type(mean))

38
bindings/python/cntk/ops/functions.py
Просмотреть файл

@ -5,7 +5,7 @@ from enum import Enum, unique
@unique
class CloneMethod(Enum):
'''
Describes different ways how :class:`~cntk.ops.functions.Function.forward`
Describes different ways how :func:`~cntk.ops.functions.Function.clone`
works.
'''
@ -125,11 +125,11 @@ class Function(cntk_py.Function):
Args:
method (:class:`CloneMethod`): one of
* 'cloned': the returned function gets its own copy of parameters (default)
* 'shared': the returned function shares its parameters with this function
* 'constant': parameters are cloned and made immutable (constant).
* 'clone': the returned function gets its own copy of parameters (default)
* 'share': the returned function shares its parameters with this function
* 'freeze': parameters are cloned and made immutable (constant).
substitutions (`dict`): a dictionary mapping variables in this
substitutions (dict): a dictionary mapping variables in this
function to variables in the cloned function
Returns:
@ -157,9 +157,9 @@ class Function(cntk_py.Function):
arguments: maps variables to their input data. The interpretation depends on
the input type:
* `dict`: keys are input variable or names, and values are the input data.
* dict: keys are input variable or names, and values are the input data.
* any other type: if node has an unique input, ``arguments`` is mapped to this input.
For nodes with more than one input, only `dict` is allowed.
For nodes with more than one input, only dict is allowed.
In both cases, every every sample in the data will be interpreted
as a new sequence. To mark samples as continuations of the
previous sequence, specify ``arguments`` as `tuple`: the
@ -208,9 +208,9 @@ class Function(cntk_py.Function):
arguments: maps variables to their
input data. The interpretation depends on the input type:
* `dict`: keys are input variable or names, and values are the input data.
* dict: keys are input variable or names, and values are the input data.
* any other type: if node has an unique input, ``arguments`` is mapped to this input.
For nodes with more than one input, only `dict` is allowed.
For nodes with more than one input, only dict is allowed.
In both cases, every every sample in the data will be interpreted
as a new sequence. To mark samples as continuations of the
previous sequence, specify ``arguments`` as ``tuple``: the
@ -220,7 +220,7 @@ class Function(cntk_py.Function):
Data should be either NumPy arrays or a
:class:`~cntk.io.MinibatchData` instance.
outputs (iterable): outputs to fetch values for.
keep_for_backward (`set`, default `None`): the subset of the
keep_for_backward (set, default `None`): the subset of the
Function's output variables for which gradients shall be calculated
in a subsequent backward call. If `None`, the returned state will
be `None` and a subsequent call to :func:`backward` will not be
@ -230,7 +230,7 @@ class Function(cntk_py.Function):
computation is. If `None`, the default device is used.
Returns:
A tuple (`BackpropState`, `map` of outputs to NumPy arrays). The
A tuple (BackpropState, map of outputs to NumPy arrays). The
BackpropState is a handle taken by :func:`backward`.
'''
if device is None:
@ -270,15 +270,15 @@ class Function(cntk_py.Function):
array([[[ 0.25]]], dtype=float32)
Args:
state (`BackPropState`): state obtained from a previous call to the
state (BackPropState): state obtained from a previous call to the
func:`cntk.ops.Function.forward` method on this Function for the
computation that this gradient backpropagation corresponds to.
root_gradients (`dict`): the gradients that will be backpropagated
variables (`set`): a list of input variables with respect to which
root_gradients (dict): the gradients that will be backpropagated
variables (set): a list of input variables with respect to which
the gradients have to be computed.
Returns:
`dict`: mapping of ``variables`` to NumPy arrays
dict: mapping of ``variables`` to NumPy arrays
'''
root_gradients = sanitize_var_map(self.outputs, root_gradients)
@ -353,7 +353,7 @@ class Function(cntk_py.Function):
specified replacements in the map.
Args:
substitutions (``dict``): map from placeholder to variables
substitutions (dict): map from placeholder to variables
Returns:
:class:`Function`: itself
@ -383,8 +383,8 @@ class Function(cntk_py.Function):
Save this function graph into a model file
Args:
filename (`str`): model path
use_legacy_format (`str`): if 'True', model is stored using legacy format.
filename (str): model path
use_legacy_format (str): if 'True', model is stored using legacy format.
Otherwise, it's stored using protobuf-based protocol serialization.
'''
return super(Function, self).save_model(filename, use_legacy_format)
@ -395,7 +395,7 @@ class Function(cntk_py.Function):
Restore the models parameters from a saved model file
Args:
filename (`str`): saved model path
filename (str): saved model path
Returns:
`None`: this method only has the side-effect of loading the model parameters from the file

20
bindings/python/cntk/ops/sequence/__init__.py
Просмотреть файл

@ -30,7 +30,7 @@ def is_first(seq, name=''):
name (str): the name of the node in the network
Returns:
:class:`cntk.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import is_first
seq = sanitize_input(seq, get_data_type(seq))
@ -57,7 +57,7 @@ def is_last(seq, name=''):
name (str): the name of the node in the network
Returns:
:class:`cntk.Function`:
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import is_last
seq = sanitize_input(seq, get_data_type(seq))
@ -80,7 +80,7 @@ def slice(seq, begin_index, end_index, name=''):
Indexing in NumPy: http://docs.scipy.org/doc/numpy/reference/arrays.indexing.html
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import sequence_slice
seq = sanitize_input(seq, get_data_type(seq))
@ -106,7 +106,7 @@ def first(seq, name=''):
seq: the symbolic tensor denoting a sequence
name (str): the name of the node in the network
Returns:
:class:`cntk.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import first
seq = sanitize_input(seq, get_data_type(seq))
@ -134,7 +134,7 @@ def last(seq, name=''):
name (str): the name of the node in the network
Returns:
:class:`cntk.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import last
seq = sanitize_input(seq, get_data_type(seq))
@ -164,7 +164,7 @@ def where(condition, name=''):
name (str): the name of the node in the network
Returns:
:class:`cntk.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import where
condition = sanitize_input(condition, get_data_type(condition))
@ -199,7 +199,7 @@ def gather(seq, condition, name=''):
elements should be selected
name (str): the name of the node in the network
Returns:
:class:`cntk.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import gather
seq = sanitize_input(seq, get_data_type(seq))
@ -249,7 +249,7 @@ def scatter(seq, condition, name=''):
elements should be copied
name (str): the name of the node in the network
Returns:
:class:`cntk.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import scatter
seq = sanitize_input(seq, get_data_type(seq))
@ -296,7 +296,7 @@ def broadcast_as(operand, broadcast_as_operand, name=''):
name (str): the name of the node in the network
Returns:
:class:`cntk.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import broadcast_as
operand = sanitize_input(operand, get_data_type(operand))
@ -316,7 +316,7 @@ def reduce_sum(seq, name=''):
name (`str`, optional): the name of the Function instance in the network
Returns:
:class:`cntk.ops.functions.Function`
:class:`~cntk.ops.functions.Function`
'''
from cntk.cntk_py import sequence_reduce_sum
seq = sanitize_input(seq, get_data_type(seq))

12
bindings/python/cntk/persist.py
Просмотреть файл

@ -13,9 +13,9 @@ def save_model(root_op, filename, use_legacy_format=True):
Save the network of ``root_op`` in ``filename``.
Args:
root_op (:class:`cntk.functions.Function`): op of the graph to save
filename (`str`): filename to store the model in
use_legacy_format (`str`): if 'True', model is stored using legacy format.
root_op (:class:`~cntk.functions.Function`): op of the graph to save
filename (str): filename to store the model in
use_legacy_format (str): if 'True', model is stored using legacy format.
Otherwise, it's stored using protobuf-based protocol serialization.
'''
root_op.save_model(filename, use_legacy_format)
@ -27,10 +27,10 @@ def load_model(filename, dtype=np.float32, device=None):
`:func:save_model`.
Args:
filename (`str`): filename to load the model from
dtype ('float', 'double', or NumPy type, default ``np.float32``): data
filename (str): filename to load the model from
dtype ('float', 'double', or NumPy type, default np.float32): data
type of the operation
device (:class:`cntk.DeviceDescriptor`, default is the default device):
device (:class:`~cntk.DeviceDescriptor`, default is the default device):
instance of DeviceDescriptor
Returns:

8
bindings/python/cntk/trainer.py
Просмотреть файл

@ -26,7 +26,7 @@ class Trainer(cntk_py.Trainer):
model (:class:`~cntk.ops.functions.Function`): root node of the function to train
loss_function (:class:`~cntk.ops.functions.Function`): loss function
eval_function (:class:`~cntk.ops.functions.Function`): evaluation function
parameter_learners (`list`): list of learners from :mod:`cntk.learner`
parameter_learners (list): list of learners from :mod:`cntk.learner`
distributed_trainer (:class:`~cntk.distributed.distributed_trainer`): distributed trainer
'''
def __init__(self, model, loss_function, eval_function, parameter_learners, distributed_trainer=None):
@ -133,8 +133,8 @@ class Trainer(cntk_py.Trainer):
specified file location.
Args:
filename (`str`): filename to store the checkpoint
use_legacy_format (`str`): if 'True', model is stored using legacy format.
filename (str): filename to store the checkpoint
use_legacy_format (str): if 'True', model is stored using legacy format.
Otherwise, it's stored using protobuf-based protocol serialization.
'''
@ -146,7 +146,7 @@ class Trainer(cntk_py.Trainer):
specified file location.
Args:
filename (`str`): filename to restore the checkpoint from
filename (str): filename to restore the checkpoint from
'''
super(Trainer, self).restore_from_checkpoint(filename)