def __init__(self,
axis=-1,
momentum=0.99,
epsilon=1e-3,
center=True,
scale=True,
beta_initializer='zeros',
gamma_initializer='ones',
moving_mean_initializer='zeros',
moving_variance_initializer='ones',
beta_regularizer=None,
gamma_regularizer=None,
name=None,
**kwargs):
"""Create a ``BatchNormalization`` layer.
Parameters
----------
axis : int, optional, default=-1
The channel axis.
momentum : float, optional, default=0.99
The decay factor of running average.
epsilon : float, optional, default=1e-3
The epsilon value.
center : bool, optional, default=True
**False** to freeze the ``beta`` anyway.
scale : bool, optional, default=True
**False** to freeze the ``gamma`` anyway.
beta_initializer : Union[callable, str], optional
The initializer for beta tensor.
gamma_initializer : Union[callable, str], optional
The initializer for gamma tensor.
moving_mean_initializer : Union[callable, str], optional
The initializer for moving mean tensor.
moving_variance_initializer : Union[callable, str], optional
The initializer for moving variance tensor.
beta_regularizer : Union[callable, str], optional
The regularizer for beta tensor.
gamma_regularizer : Union[callable, str], optional
The regularizer for gamma tensor.
"""
super(BatchNormalization, self).__init__(name=name, **kwargs)
self.axis = axis
self.momentum = momentum
self.epsilon = epsilon
self.center = center
self.scale = scale
self.beta_initializer = initializers.get(beta_initializer)
self.gamma_initializer = initializers.get(gamma_initializer)
self.moving_mean_initializer = initializers.get(
moving_mean_initializer)
self.moving_variance_initializer = initializers.get(
moving_variance_initializer)
self.beta_regularizer = regularizers.get(beta_regularizer)
self.gamma_regularizer = regularizers.get(gamma_regularizer)
self.beta = None
self.gamma = None
self.moving_mean = None
self.moving_variance = None
def __init__(self,
filters,
kernel_size,
strides=1,
padding='valid',
data_format='channels_last',
dilation_rate=1,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
**kwargs):
"""Create a ``Conv3D`` Layer.
Parameters
----------
filters : int
The number of output filters.
kernel_size : Union[int, Sequence[int]]
The shape of convolution window.
strides : Union[int, Sequence[int]], optional, default=1
The stride of convolution window.
padding : Union[str, Sequence[int]], optional
The padding algorithm or size.
data_format : str, optional, default='channels_last'
``'channels_first'`` or ``'channels_last'``.
dilation_rate : Union[int, Sequence[int]], optional, default=1
The rate of dilated convolution.
activation : Union[callable, str], optional
The optional activation function.
use_bias : bool, optional, default=True
Add a bias tensor to output or not.
kernel_initializer : Union[callable, str], optional
The initializer for kernel tensor.
bias_initializer : Union[callable, str], optional
The initializer for bias tensor.
kernel_regularizer : Union[callable, str], optional
The regularizer for kernel tensor.
bias_regularizer : Union[callable, str], optional
The regularizer for bias tensor.
"""
super(Conv3D, self).__init__(
rank=3,
filters=filters,
kernel_size=kernel_size,
strides=strides,
padding=padding,
data_format=data_format,
dilation_rate=dilation_rate,
activation=activations.get(activation),
use_bias=use_bias,
kernel_initializer=initializers.get(kernel_initializer),
bias_initializer=initializers.get(bias_initializer),
kernel_regularizer=regularizers.get(kernel_regularizer),
bias_regularizer=regularizers.get(bias_regularizer),
**kwargs)
def __init__(self,
units,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
**kwargs):
"""Create a ``Dense`` layer.
Parameters
----------
units : int
The number of output units.
activation : Union[callable, str], optional
The optional activation function.
use_bias : bool, optional, default=True
``True`` to apply a ``bias``.
kernel_initializer : Union[callable, str], optional
The initializer for kernel tensor.
bias_initializer : Union[callable, str], optional
The initializer for bias tensor.
kernel_regularizer : Union[callable, str], optional
The regularizer for kernel tensor.
bias_regularizer : Union[callable, str], optional
The regularizer for bias tensor.
"""
super(Dense, self).__init__(**kwargs)
self.input_dim = kwargs.get('input_dim', None)
self.units = int(units)
self.activation = activations.get(activation)
self.use_bias = use_bias
self.kernel_initializer = initializers.get(kernel_initializer)
self.bias_initializer = initializers.get(bias_initializer)
self.kernel_regularizer = regularizers.get(kernel_regularizer)
self.bias_regularizer = regularizers.get(bias_regularizer)
self.input_spec = InputSpec(min_ndim=2)
self.kernel = None
self.bias = None
def _create_weight(
name,
shape,
dtype=None,
initializer='zeros',
trainable=None,
):
if isinstance(initializer, six.string_types) or callable(initializer):
initializer = initializers.get(initializer)
return variables.get_variable(
name=name,
shape=shape,
initializer=initializer,
dtype=dtype if dtype is not None else dtypes.float32,
trainable=trainable if trainable is not None else True,
use_resource=True,
)
def add_weight(self,
name=None,
shape=None,
dtype=None,
initializer=None,
regularizer=None,
trainable=True,
use_resource=None,
**kwargs):
"""Add a new variable as the weight.
Parameters
----------
name : str, optional
The optional variable name.
shape : Sequence[int], optional
The variable shape.
dtype : str, optional
The optional data type.
initializer : Union[callable, str], optional
The optional initializer.
regularizer : Union[callable, str], optional
The optional regularizer.
trainable : bool, optional, default=True
``True`` to add to the ``trainable`` collection.
use_resource : bool, optional, default=True
``True`` to set as a ``ResourceVariable``.
"""
if shape is None:
shape = ()
initializer = initializers.get(initializer)
regularizer = regularizers.get(regularizer)
# Determine the data type
if dtype is None:
dtype = self.dtype or dtypes.float32
dtype = dtypes.as_dtype(dtype)
# Determine the variable flags
trainable = True if trainable is None else trainable
use_resource = True if use_resource is None else use_resource
# Determine the initializer
if initializer is None:
if dtype.is_floating:
initializer = initializers.glorot_uniform()
elif dtype.is_integer or dtype.is_unsigned or dtype.is_bool:
initializer = initializers.zeros()
else:
raise ValueError('Excepted an initializer set for variable')
variable = tf_variables.get_variable(
name=name,
shape=shape,
initializer=initializer,
regularizer=regularizer,
dtype=dtype,
trainable=trainable,
use_resource=use_resource,
)
if trainable:
self._trainable_weights.append(variable)
else:
self._non_trainable_weights.append(variable)
return variable
def __init__(
self,
rank,
filters,
kernel_size,
strides=1,
padding='valid',
data_format='channels_last',
dilation_rate=1,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
trainable=True,
name=None,
**kwargs,
):
"""Create a ``Conv`` Layer.
Parameters
----------
rank : int
The number of spatial axes.
filters : int, optional
The number of output filters.
kernel_size : Union[int, Sequence[int]]
The shape of convolution window.
strides : Union[int, Sequence[int]], optional, default=1
The stride of convolution window.
padding : Union[int, Sequence[int], str], optional
The padding algorithm or size.
data_format : str, optional, default='channels_last'
``'channels_first'`` or ``'channels_last'``.
dilation_rate : Union[int, Sequence[int]], optional, default=1
The rate of dilated convolution.
activation : Union[callable, str], optional
The optional activation function.
use_bias : bool, optional, default=True
Add a bias tensor to output or not.
kernel_initializer : Union[callable, str], optional
The initializer for kernel tensor.
bias_initializer : Union[callable, str], optional
The initializer for bias tensor.
kernel_regularizer : Union[callable, str], optional
The regularizer for kernel tensor.
bias_regularizer : Union[callable, str], optional
The regularizer for bias tensor.
"""
super(Conv, self).__init__(trainable=trainable, name=name, **kwargs)
self.rank = rank
self.filters = filters
self.kernel_size = conv_utils.normalize_tuple(kernel_size, rank)
self.strides = conv_utils.normalize_tuple(strides, rank)
self.padding = conv_utils.normalize_padding(padding)
self.data_format = conv_utils.normalize_data_format(data_format)
self.dilation_rate = conv_utils.normalize_tuple(dilation_rate, rank)
self.activation = activations.get(activation)
self.use_bias = use_bias
self.kernel_initializer = initializers.get(kernel_initializer)
self.bias_initializer = initializers.get(bias_initializer)
self.kernel_regularizer = regularizers.get(kernel_regularizer)
self.bias_regularizer = regularizers.get(bias_regularizer)
self.input_spec = InputSpec(ndim=self.rank + 2)
self.conv_function = kwargs.get('conv_function', nn_ops.convolution)
self.kernel = None
self.bias = None
