def build(self, input_shape):
if self.data_format == 'channels_first':
input_dim, input_length = input_shape[1], input_shape[2]
else:
input_dim, input_length = input_shape[2], input_shape[1]
if input_dim is None:
raise ValueError(
'Axis 2 of input should be fully-defined. '
'Found shape:', input_shape)
self.output_length = conv_utils.conv_output_length(
input_length, self.kernel_size[0], self.padding, self.strides[0])
self.kernel_shape = (self.output_length,
self.kernel_size[0] * input_dim, self.filters)
self.kernel = self.add_weight(shape=self.kernel_shape,
initializer=self.kernel_initializer,
name='kernel',
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint)
if self.use_bias:
self.bias = self.add_weight(shape=(self.output_length,
self.filters),
initializer=self.bias_initializer,
name='bias',
regularizer=self.bias_regularizer,
constraint=self.bias_constraint)
else:
self.bias = None
if self.data_format == 'channels_first':
self.input_spec = InputSpec(ndim=3, axes={1: input_dim})
else:
self.input_spec = InputSpec(ndim=3, axes={-1: input_dim})
self.built = True
def __init__(self,
return_sequences=False,
return_state=False,
go_backwards=False,
stateful=False,
time_major=False,
**kwargs):
# We invoke the base layer's initializer directly here because we do not
# want to create RNN cell instance.
super(RNN, self).__init__(**kwargs) # pylint: disable=bad-super-call
self.return_sequences = return_sequences
self.return_state = return_state
self.go_backwards = go_backwards
self.stateful = stateful
self.time_major = time_major
self.supports_masking = False
self.input_spec = [InputSpec(ndim=3)]
if hasattr(self.cell.state_size, '__len__'):
state_size = self.cell.state_size
else:
state_size = [self.cell.state_size]
self.state_spec = [InputSpec(shape=(None, dim)) for dim in state_size]
self.constants_spec = None
self._states = None
self._num_constants = None
self._num_inputs = None
self._vector_shape = constant_op.constant([-1])
def __init__(self,
feature_units,
activation="relu",
return_attention=False,
node_axis="row",
merge_method="add",
use_attention_kernel=True,
**kwargs):
super(SimpleAttentionLayer, self).__init__(units=feature_units,
activation=activation,
**kwargs)
if merge_method == "concat" and not use_attention_kernel:
raise Exception("Can't use concat without attention")
self.return_attention = return_attention
self.node_axis = node_axis
self.merge_method = merge_method
self.use_attention_kernel = use_attention_kernel
self.input_spec = [InputSpec(ndim=3), InputSpec(ndim=3)]
self.supports_masking = False
self.self_kernel = None
self.neighbor_kernel = None
self.attention_kernel = None
self.bias = None
def build(self, input_shape):
# Note input_shape will be list of shapes of initial states and
# constants if these are passed in __call__.
if self._num_constants is not None:
constants_shape = input_shape[-self._num_constants:] # pylint: disable=E1130
else:
constants_shape = None
if isinstance(input_shape, list):
input_shape = input_shape[0]
batch_size = input_shape[0] if self.stateful else None
self.input_spec[0] = InputSpec(shape=(batch_size, None) +
input_shape[2:5])
# allow cell (if layer) to build before we set or validate state_spec
if isinstance(self.cell, Layer):
step_input_shape = (input_shape[0], ) + input_shape[2:]
if constants_shape is not None:
self.cell.build([step_input_shape] + constants_shape)
else:
self.cell.build(step_input_shape)
# set or validate state_spec
if hasattr(self.cell.state_size, '__len__'):
state_size = list(self.cell.state_size)
else:
state_size = [self.cell.state_size]
if self.state_spec is not None:
# initial_state was passed in call, check compatibility
if self.cell.data_format == 'channels_first':
ch_dim = 1
elif self.cell.data_format == 'channels_last':
ch_dim = 3
if [spec.shape[ch_dim] for spec in self.state_spec] != state_size:
raise ValueError(
'An initial_state was passed that is not compatible with '
'`cell.state_size`. Received `state_spec`={}; '
'However `cell.state_size` is '
'{}'.format([spec.shape for spec in self.state_spec],
self.cell.state_size))
else:
if self.cell.data_format == 'channels_first':
self.state_spec = [
InputSpec(shape=(None, dim, None, None))
for dim in state_size
]
elif self.cell.data_format == 'channels_last':
self.state_spec = [
InputSpec(shape=(None, None, None, dim))
for dim in state_size
]
if self.stateful:
self.reset_states()
self.built = True
def __init__(self,
units,
kernel_initializer='glorot_uniform',
recurrent_initializer='orthogonal',
bias_initializer='zeros',
unit_forget_bias=True,
kernel_regularizer=None,
recurrent_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
recurrent_constraint=None,
bias_constraint=None,
return_sequences=False,
return_state=False,
go_backwards=False,
stateful=False,
time_major=False,
**kwargs):
super(RNN, self).__init__(**kwargs) # pylint: disable=bad-super-call
self.units = units
cell_spec = collections.namedtuple('cell', ['state_size', 'output_size'])
self.cell = cell_spec(
state_size=(self.units, self.units), output_size=self.units)
self.kernel_initializer = initializers.get(kernel_initializer)
self.recurrent_initializer = initializers.get(recurrent_initializer)
self.bias_initializer = initializers.get(bias_initializer)
self.unit_forget_bias = unit_forget_bias
self.kernel_regularizer = regularizers.get(kernel_regularizer)
self.recurrent_regularizer = regularizers.get(recurrent_regularizer)
self.bias_regularizer = regularizers.get(bias_regularizer)
self.activity_regularizer = regularizers.get(activity_regularizer)
self.kernel_constraint = constraints.get(kernel_constraint)
self.recurrent_constraint = constraints.get(recurrent_constraint)
self.bias_constraint = constraints.get(bias_constraint)
self.return_sequences = return_sequences
self.return_state = return_state
self.go_backwards = go_backwards
self.stateful = stateful
self.time_major = time_major
self._num_constants = None
self._num_inputs = None
self._states = None
self.input_spec = [InputSpec(ndim=3)]
self.state_spec = [
InputSpec(shape=(None, dim)) for dim in (self.units, self.units)
]
def __init__(
self,
feature_units,
attn_heads=1,
attn_heads_reduction="concat", # {"concat", "average"}
dropout_rate=0.5,
activation="relu",
attn_kernel_initializer="glorot_uniform",
attn_kernel_regularizer=None,
attn_kernel_constraint=None,
attention=True,
return_attention=False,
**kwargs):
if attn_heads_reduction not in {"concat", "average"}:
raise ValueError("Possbile reduction methods: concat, average")
super(GraphAttentionLayer, self).__init__(units=feature_units,
activation=activation,
**kwargs)
# Number of attention heads (K in the paper)
self.attn_heads = attn_heads
# Eq. 5 and 6 in the paper
self.attn_heads_reduction = attn_heads_reduction
# Internal dropout rate
self.dropout_rate = dropout_rate
self.attn_kernel_initializer \
= initializers.get(attn_kernel_initializer)
self.attn_kernel_regularizer \
= regularizers.get(attn_kernel_regularizer)
self.attn_kernel_constraint = constraints.get(attn_kernel_constraint)
self.attention = attention
self.return_attention = return_attention
self.input_spec = [InputSpec(ndim=3), InputSpec(ndim=3)]
self.supports_masking = False
# Populated by build()
self.kernels = []
self.biases = []
self.neighbor_kernels = []
self.attn_kernels = []
self.attention_biases = []
if attn_heads_reduction == "concat":
# Output will have shape (..., K * F")
self.output_dim = self.units * self.attn_heads
else:
# Output will have shape (..., F")
self.output_dim = self.units
def __call__(self, inputs, initial_state=None, constants=None, **kwargs):
inputs, initial_state, constants = _standardize_args(
inputs, initial_state, constants, self._num_constants)
if initial_state is None and constants is None:
return super(STLSTM2D, self).__call__(inputs, **kwargs)
# If any of `initial_state` or `constants` are specified and are Keras
# tensors, then add them to the inputs and temporarily modify the
# input_spec to include them.
additional_inputs = []
additional_specs = []
if initial_state is not None:
kwargs['initial_state'] = initial_state
additional_inputs += initial_state
self.state_spec = []
for state in initial_state:
shape = K.int_shape(state)
self.state_spec.append(InputSpec(shape=shape))
additional_specs += self.state_spec
if constants is not None:
kwargs['constants'] = constants
additional_inputs += constants
self.constants_spec = [
InputSpec(shape=K.int_shape(constant))
for constant in constants
]
self._num_constants = len(constants)
additional_specs += self.constants_spec
# at this point additional_inputs cannot be empty
for tensor in additional_inputs:
if K.is_keras_tensor(tensor) != K.is_keras_tensor(
additional_inputs[0]):
raise ValueError('The initial state or constants of an RNN'
' layer cannot be specified with a mix of'
' Keras tensors and non-Keras tensors')
if K.is_keras_tensor(additional_inputs[0]):
# Compute the full input spec, including state and constants
full_input = [inputs] + additional_inputs
full_input_spec = self.input_spec + additional_specs
# Perform the call with temporarily replaced input_spec
original_input_spec = self.input_spec
self.input_spec = full_input_spec
output = super(STLSTM2D, self).__call__(full_input, **kwargs)
self.input_spec = original_input_spec
return output
else:
return super(STLSTM2D, self).__call__(inputs, **kwargs)
def build(self, input_shape):
if self.data_format == 'channels_first':
input_dim, input_length = input_shape[1], input_shape[2]
else:
input_dim, input_length = input_shape[2], input_shape[1]
if input_dim is None:
raise ValueError(
'Axis 2 of input should be fully-defined. '
'Found shape:', input_shape)
self.output_length = self.mask.shape[1]
# print("output length is " + str(self.output_length))
if self.data_format == 'channels_first':
self.kernel_shape = (input_dim, input_length, self.filters,
self.output_length)
else:
self.kernel_shape = (input_length, input_dim, self.output_length,
self.filters)
self.kernel = self.add_weight(
shape=(len(self.mask.data),
), # sum of all nonzero values in mask sum(sum(mask))
initializer=self.kernel_initializer,
name='kernel',
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint)
self.kernel_mask = get_locallyDirected1D_mask(
self.mask,
self.kernel,
data_format=self.data_format,
dtype=self.kernel.dtype)
if self.use_bias:
self.bias = self.add_weight(shape=(self.output_length,
self.filters),
initializer=self.bias_initializer,
name='bias',
regularizer=self.bias_regularizer,
constraint=self.bias_constraint)
else:
self.bias = None
if self.data_format == 'channels_first':
self.input_spec = InputSpec(ndim=3, axes={1: input_dim})
else:
self.input_spec = InputSpec(ndim=3, axes={-1: input_dim})
self.built = True
def build(self, input_shape):
"""
Reimplementation of tf.keras.layers.Conv3D creating multiple conv ops
with varying degree of dilation with shared weights
"""
input_shape = tensor_shape.TensorShape(input_shape)
if self.data_format == 'channels_first':
self.channel_axis = 1
else:
self.channel_axis = -1
if input_shape[self.channel_axis].value is None:
raise ValueError('The channel dimension of the inputs '
'should be defined. Found `None`.')
input_dim = int(input_shape[self.channel_axis])
self.input_spec = InputSpec(ndim=self.rank + 2,
axes={self.channel_axis: input_dim})
# Prepare weights and conv ops for dilated conv layers
with tf.name_scope("dilated_conv_weights"):
self.build_dilated_conv(input_shape, input_dim)
# Prepare weights and conv ops for attention mechanism
with tf.name_scope("attention_weights"):
self.build_attention(input_shape, input_dim)
# Set build flag, needed if called directly
self.built = True
def build(self, input_shape):
dtype = dtypes.as_dtype(self.dtype or K.floatx())
if not (dtype.is_floating or dtype.is_complex):
raise TypeError(
'Unable to build `Dense` layer with non-floating point '
'dtype %s' % (dtype, ))
input_shape = tensor_shape.TensorShape(input_shape)
if input_shape[-1].value is None:
raise ValueError('The last dimension of the inputs to `Dense` '
'should be defined. Found `None`.')
self.input_spec = InputSpec(min_ndim=2,
axes={-1: input_shape[-1].value})
self.kernel = self.add_weight(
'kernel',
shape=[input_shape[-1].value, self.units],
initializer=self.kernel_initializer,
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
dtype=self.dtype,
trainable=True)
if self.use_bias:
self.bias = self.add_weight('bias',
shape=[
self.units,
],
initializer=self.bias_initializer,
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
dtype=self.dtype,
trainable=True)
else:
self.bias = None
self.built = True
def __init__(self,
units,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
bias_constraint=None,
**kwargs):
super(GraphConvolutionalLayer, self).__init__(
activity_regularizer=regularizers.get(activity_regularizer),
**kwargs)
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.kernel_constraint = constraints.get(kernel_constraint)
self.bias_constraint = constraints.get(bias_constraint)
self.input_spec = InputSpec(min_ndim=3)
def __init__(self,
cell,
return_sequences=False,
return_state=False,
go_backwards=False,
stateful=False,
unroll=False,
**kwargs):
if unroll:
raise TypeError('Unrolling isn\'t possible with '
'convolutional RNNs.')
if isinstance(cell, (list, tuple)):
# The StackedConvRNN2DCells isn't implemented yet.
raise TypeError('It is not possible at the moment to'
'stack convolutional cells.')
super(ConvRNN2D, self).__init__(cell,
return_sequences,
return_state,
go_backwards,
stateful,
unroll,
**kwargs)
self.input_spec = [InputSpec(ndim=5)]
self.states = None
self._num_constants = None
def __init__(self, output_dim, init='glorot_uniform',
init_sigma=lambda shape, name:init_uniform(shape, -10, -5, name=name),
activation='linear', weights=None,
W_regularizer=None, b_regularizer=None, W_sigma_regularizer=None, b_sigma_regularizer=None,
activity_regularizer=None,
bias=True, input_dim=None, **kwargs):
self.init = initializers.get(init)
self.init_sigma = init_sigma
self.activation = activations.get(activation)
self.output_dim = output_dim
self.input_dim = input_dim
self.uses_learning_phase = True
self.W_regularizer = regularizers.get(W_regularizer)
self.b_regularizer = regularizers.get(b_regularizer)
self.W_sigma_regularizer = regularizers.get(W_sigma_regularizer)
self.b_sigma_regularizer = regularizers.get(b_sigma_regularizer)
self.activity_regularizer = regularizers.get(activity_regularizer)
self.bias = bias
self.initial_weights = weights
self.input_spec = [InputSpec(ndim=2)]
if self.input_dim:
kwargs['input_shape'] = (self.input_dim,)
super(BayesianDense, self).__init__(**kwargs)
def __init__(self,
units,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
bias_constraint=None,
**kwargs):
if 'input_shape' not in kwargs and 'input_dim' in kwargs:
kwargs['input_shape'] = (kwargs.pop('input_dim'),)
super(Dense, self).__init__(
activity_regularizer=regularizers.get(activity_regularizer), **kwargs)
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.kernel_constraint = constraints.get(kernel_constraint)
self.bias_constraint = constraints.get(bias_constraint)
self.supports_masking = True
self.input_spec = InputSpec(min_ndim=2)
def build(self, input_shape):
input_shape = tensor_shape.TensorShape(input_shape)
if input_shape[-1].value is None:
raise ValueError('The last dimension of the inputs to `Dense` '
'should be defined. Found `None`.')
self.input_spec = InputSpec(min_ndim=2,
axes={-1: input_shape[-1].value})
self.kernel = self.add_weight(
'kernel',
shape=[input_shape[-1].value, self.units],
initializer=self.kernel_initializer,
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
dtype=self.dtype,
trainable=True)
if self.use_bias:
self.bias = self.add_weight(
'bias',
shape=[self.units,],
initializer=self.bias_initializer,
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
dtype=self.dtype,
trainable=True)
else:
self.bias = None
self.built = True
def __init__(self,
tied_layer='',
activation=None,
use_bias=True,
bias_initializer='zeros',
bias_regularizer=None,
activity_regularizer=None,
bias_constraint=None,
varName='',
varShape=[],
**kwargs):
if 'input_shape' not in kwargs and 'input_dim' in kwargs:
kwargs['input_shape'] = (kwargs.pop('input_dim'), )
super(DenseTied, self).__init__(
activity_regularizer=regularizers.get(activity_regularizer),
**kwargs)
if tied_layer != '':
self.kernelFrom = tied_layer.kernel.name
self.varName = varName
self.varShape = varShape
self.activation = activations.get(activation)
self.use_bias = use_bias
self.bias_initializer = initializers.get(bias_initializer)
self.bias_regularizer = regularizers.get(bias_regularizer)
self.bias_constraint = constraints.get(bias_constraint)
self.supports_masking = True
self.input_spec = InputSpec(min_ndim=2)
def __init__(self,
num_units,
activation=None,
reuse=None,
kernel_initialier=None,
bias_initializer=None,
name=None,
dtype=None,
**kwargs):
super(GRUCell, self).__init__(_reuse=reuse,
name=name,
dtype=dtype,
**kwargs)
#_check_supported_dtypes(self.dtype)
if context.executing_eagerly() and context.num_gpus() > 0:
logging.warn("This is not optimized for performance.")
self.input_spec = InputSpec(ndim=2)
self._num_units = num_units
if activation:
self._activation = activations.get(activation)
else:
self._activation = math_ops.tanh
self._kernel_initializer = initializers.get(kernel_initialier)
self._bias_initializer = initializers.get(bias_initializer)
def __init__(
self,
units,
activation=None,
use_bias=True,
# kernel_initializer='glorot_uniform',
bias_initializer='zeros',
# kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
# kernel_constraint=None,
bias_constraint=None,
tied_to=None,
**kwargs):
if 'input_shape' not in kwargs and 'input_dim' in kwargs:
kwargs['input_shape'] = (kwargs.pop('input_dim'), )
super(DenseTied, self).__init__(
activity_regularizer=regularizers.get(activity_regularizer),
**kwargs)
self.tied_to = tied_to
self.units = int(units)
self.activation = activations.get(activation)
"""transposed weights are variables and don't use any regularizators"""
# self.kernel_initializer = None
# self.kernel_constraint = None #regularizers.get(kernel_regularizer)
# self.kernel_regularizer = None #constraints.get(kernel_constraint)
"""biases are still initialized and regularized"""
self.use_bias = use_bias
self.bias_initializer = initializers.get(bias_initializer)
self.bias_regularizer = regularizers.get(bias_regularizer)
self.bias_constraint = constraints.get(bias_constraint)
self.supports_masking = True
self.input_spec = InputSpec(min_ndim=2)
def build(self, input_shape):
ndim = len(input_shape)
if self.axis == 0:
raise ValueError('Axis cannot be zero')
if (self.axis is not None) and (ndim == 2):
raise ValueError('Cannot specify axis for rank 1 tensor')
self.input_spec = InputSpec(ndim=ndim)
if self.axis is None:
shape = (1, )
else:
shape = (input_shape[self.axis], )
if self.scale:
self.gamma = self.add_weight(shape=shape,
name='gamma',
initializer=self.gamma_initializer,
regularizer=self.gamma_regularizer,
constraint=self.gamma_constraint)
else:
self.gamma = None
if self.center:
self.beta = self.add_weight(shape=shape,
name='beta',
initializer=self.beta_initializer,
regularizer=self.beta_regularizer,
constraint=self.beta_constraint)
else:
self.beta = None
self.built = True
def build(self, input_shape):
input_shape = tensor_shape.TensorShape(input_shape)
if len(input_shape) != 4:
raise ValueError('Inputs should have rank 4. Received input '
'shape: ' + str(input_shape))
channel_axis = self._get_channel_axis()
if input_shape.dims[channel_axis].value is None:
raise ValueError('The channel dimension of the inputs '
'should be defined. Found `None`.')
input_dim = int(input_shape[channel_axis])
self.input_spec = InputSpec(ndim=4, axes={channel_axis: input_dim})
kernel_shape = self.kernel_size + (self.filters, input_dim)
if self.tied_to is not None:
self.kernel = self.tied_to.kernel
self._non_trainable_weights.append(self.kernel)
else:
self.kernel = self.add_weight(name='kernel',
shape=kernel_shape,
initializer=self.kernel_initializer,
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
trainable=True,
dtype=self.dtype)
if self.use_bias:
self.bias = self.add_weight(name='bias',
shape=(self.filters, ),
initializer=self.bias_initializer,
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
trainable=True,
dtype=self.dtype)
else:
self.bias = None
self.built = True
def __init__(self, kernel_size=3, activation='elu'):
super().__init__()
self.kernel_size = kernel_size
self.activation = activation
self._input_spec = InputSpec(ndim=3)
self.supports_masking = True
def build(self, input_shape):
input_shape = tensor_shape.TensorShape(
input_shape) # The input should be [batch_size x units x N]
if len(input_shape) != 3 or input_shape[
-1].value is None or input_shape[-2].value is None:
raise ValueError(
'The last two dimensions of the GCN are wrong: {}. Input should be [BS x N x units]'
.format(input_shape))
self.input_spec = InputSpec(min_ndim=3,
axes={
-1: input_shape[-1].value,
-2: input_shape[-2].value
})
self.kernel = self.add_weight(
'kernel',
shape=[input_shape[-1].value, self.units],
initializer=self.kernel_initializer,
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
dtype=self.dtype,
trainable=True)
if self.use_bias:
self.bias = self.add_weight('bias',
shape=[
self.units,
],
initializer=self.bias_initializer,
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
dtype=self.dtype,
trainable=True)
def __init__(self,
mask,
filters,
padding='valid',
data_format=None,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
bias_constraint=None,
**kwargs):
super(LocallyDirected1D, self).__init__(**kwargs)
self.filters = filters
self.padding = conv_utils.normalize_padding(padding)
self.data_format = conv_utils.normalize_data_format(data_format)
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.activity_regularizer = regularizers.get(activity_regularizer)
self.kernel_constraint = constraints.get(kernel_constraint)
self.bias_constraint = constraints.get(bias_constraint)
self.input_spec = InputSpec(ndim=3)
self.mask = mask
def build(self, input_shape):
"""Build `Layer`"""
input_shape = tensor_shape.TensorShape(input_shape).as_list()
self.input_spec = InputSpec(shape=input_shape)
if not self.layer.built:
self.layer.build(input_shape)
self.layer.built = False
if not hasattr(self.layer, 'kernel'):
raise ValueError('`WeightNormalization` must wrap a layer that'
' contains a `kernel` for weights')
# The kernel's filter or unit dimension is -1
self.layer_depth = int(self.layer.kernel.shape[-1])
self.norm_axes = list(range(self.layer.kernel.shape.ndims - 1))
self.layer.v = self.layer.kernel
self.layer.g = self.layer.add_variable(
name="g",
shape=(self.layer_depth, ),
initializer=initializers.get('ones'),
dtype=self.layer.kernel.dtype,
trainable=True,
aggregation=tf_variables.VariableAggregation.MEAN)
with ops.control_dependencies(
[self.layer.g.assign(self._init_norm(self.layer.v))]):
self._compute_weights()
self.layer.built = True
super(WeightNormalization, self).build()
self.built = True
def build(self, input_shape):
input_shape = tensor_shape.TensorShape(input_shape)
if self.data_format == 'channels_first':
channel_axis = 1
else:
channel_axis = -1
if input_shape.dims[channel_axis].value is None:
raise ValueError('The channel dimension of the inputs '
'should be defined. Found `None`.')
input_dim = int(input_shape[channel_axis])
kernel_shape = self.kernel_size + (input_dim, self.filters)
self.wn_g = self.add_weight(
name='wn_g',
shape=(self.filters, ),
initializer=tf.keras.initializers.RandomUniform(1, 1),
trainable=True,
dtype=self.dtype)
self.kernel = self.add_weight(name='kernel',
shape=kernel_shape,
initializer=self.kernel_initializer,
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
trainable=True,
dtype=self.dtype)
if self.use_bias:
self.bias = self.add_weight(name='bias',
shape=(self.filters, ),
initializer=self.bias_initializer,
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
trainable=True,
dtype=self.dtype)
else:
self.bias = None
self.input_spec = InputSpec(ndim=self.rank + 2,
axes={channel_axis: input_dim})
if self.padding == 'causal':
op_padding = 'valid'
else:
op_padding = self.padding
if not isinstance(op_padding, (list, tuple)):
op_padding = op_padding.upper()
self._convolution_op = nn_ops.Convolution(
input_shape,
filter_shape=self.kernel.shape,
dilation_rate=self.dilation_rate,
strides=self.strides,
padding=op_padding,
data_format=conv_utils.convert_data_format(self.data_format,
self.rank + 2))
self.built = True
def __init__(self, dims, **kwargs):
super(Permute, self).__init__(**kwargs)
self.dims = tuple(dims)
if sorted(dims) != list(range(1, len(dims) + 1)):
raise ValueError(
'Invalid permutation `dims` for Permute Layer: %s. '
'The set of indices in `dims` must be consecutive and start from 1.' %
(dims,))
self.input_spec = InputSpec(ndim=len(self.dims) + 1)
def __init__(self, rate, data_format=None, **kwargs):
super(SpatialDropout3D, self).__init__(rate, **kwargs)
if data_format is None:
data_format = K.image_data_format()
if data_format not in {'channels_last', 'channels_first'}:
raise ValueError('data_format must be in '
'{"channels_last", "channels_first"}')
self.data_format = data_format
self.input_spec = InputSpec(ndim=5)
def __init__(self, rank,
lgroups,
lfilters,
kernel_size,
strides=1,
padding='valid',
data_format=None,
dilation_rate=1,
activation=None,
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='zeros',
kernel_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
bias_constraint=None,
trainable=True,
name=None,
**kwargs):
super(_GroupConv, self).__init__(
trainable=trainable,
name=name,
activity_regularizer=regularizers.get(activity_regularizer),
**kwargs)
self.rank = rank
if rank > 2:
raise ValueError('The quick group convolution does not support 3D or any higher dimension.')
initRank = rank
self.lgroups = lgroups
self.lfilters = lfilters
self.kernel_size = conv_utils.normalize_tuple(
kernel_size, rank, 'kernel_size')
self.strides = conv_utils.normalize_tuple(strides, rank, 'strides')
self.padding = conv_utils.normalize_padding(padding)
if (self.padding == 'causal' and not isinstance(self, (Conv1D, SeparableConv1D))):
raise ValueError('Causal padding is only supported for `Conv1D` and ``SeparableConv1D`.')
self.data_format = conv_utils.normalize_data_format(data_format)
self.dilation_rate = conv_utils.normalize_tuple(
dilation_rate, rank, 'dilation_rate')
if rank == 1: # when rank=1, expand the tuples to 2D case.
self.kernel_size = (1, *self.kernel_size)
self.strides = (1, *self.strides)
self.dilation_rate = (1, *self.dilation_rate)
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.kernel_constraint = constraints.get(kernel_constraint)
self.bias_constraint = constraints.get(bias_constraint)
self.input_spec = InputSpec(ndim=self.rank + 2)
self.group_input_dim = None
self.exp_dim_pos = None
def build(self, input_shape):
if self.data_format == 'channels_last':
input_row, input_col = input_shape[1:-1]
input_filter = input_shape[3]
else:
input_row, input_col = input_shape[2:]
input_filter = input_shape[1]
if input_row is None or input_col is None:
raise ValueError('The spatial dimensions of the inputs to '
' a LocallyConnected2D layer '
'should be fully-defined, but layer received '
'the inputs shape ' + str(input_shape))
output_row = conv_utils.conv_output_length(input_row,
self.kernel_size[0],
self.padding,
self.strides[0])
output_col = conv_utils.conv_output_length(input_col,
self.kernel_size[1],
self.padding,
self.strides[1])
self.output_row = output_row
self.output_col = output_col
self.kernel_shape = (output_row * output_col, self.kernel_size[0] *
self.kernel_size[1] * input_filter, self.filters)
self.kernel = self.add_weight(shape=self.kernel_shape,
initializer=self.kernel_initializer,
name='kernel',
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint)
if self.use_bias:
self.bias = self.add_weight(shape=(output_row, output_col,
self.filters),
initializer=self.bias_initializer,
name='bias',
regularizer=self.bias_regularizer,
constraint=self.bias_constraint)
else:
self.bias = None
if self.data_format == 'channels_first':
self.input_spec = InputSpec(ndim=4, axes={1: input_filter})
else:
self.input_spec = InputSpec(ndim=4, axes={-1: input_filter})
self.built = True
def build(self, input_shape):
input_shape = tensor_shape.TensorShape(input_shape).as_list()
assert len(input_shape) >= 3
self.input_spec = InputSpec(shape=input_shape)
child_input_shape = [input_shape[0]] + input_shape[2:]
if not self.layer.built:
self.layer.build(child_input_shape)
self.layer.built = True
super(TimeDistributed, self).build()
self.built = True