def bias_initializer(_, *args, **kwargs):
return K.concatenate([
self.bias_initializer((self.filters, ), *args,
**kwargs),
initializers.Ones()((self.filters, ), *args, **kwargs),
self.bias_initializer((self.filters * 2, ), *args,
**kwargs),
initializers.Ones()((self.filters, ), *args, **kwargs),
self.bias_initializer((self.filters * 2, ), *args,
**kwargs)
])
def bias_initializer(_, *args, **kwargs):
return array_ops.concat([
self.bias_initializer((self.units * 5, ), *args, **kwargs),
initializers.Ones()((self.units, ), *args, **kwargs),
self.bias_initializer((self.units * 2, ), *args, **kwargs),
],
axis=0)
def bias_initializer(shape, *args, **kwargs):
return K.concatenate([
self.bias_initializer((self.units, ), *args, **kwargs),
initializers.Ones()((self.units, ), *args, **kwargs),
self.bias_initializer((self.units * 2, ), *args,
**kwargs),
])
def build(self, input_shape):
if type(input_shape) is list:
feature_shape = input_shape[0]
else:
feature_shape = input_shape
self.kernel_shape = (*self.kernel_size, feature_shape[-1],
self.filters)
self.kernel = self.add_weight(name='kernel',
shape=self.kernel_shape,
initializer=self.kernel_initializer,
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
trainable=True,
dtype=self.dtype)
if self.weightnorm:
self.wn_g = self.add_weight(name='wn_g',
shape=(self.filters, ),
initializer=initializers.Ones(),
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
super(Conv2D, self).build(input_shape)
def cake_bias_initializer(_, *args, **kwargs):
return K.concatenate([
self.cake_bias_initializer((self.units, ), *args,
**kwargs),
initializers.Ones()((self.units, ), *args, **kwargs),
self.cake_bias_initializer((self.units, ), *args,
**kwargs),
])
def build(self, input_shape):
super(Conv2DWeightNorm, self).build(input_shape)
self.wn_g = self.add_weight(name='wn_g',
shape=(self.filters, ),
dtype=self.dtype,
initializer=initializers.Ones(),
trainable=True)
square_sum = K.sum(K.square(self.kernel), [0, 1, 2])
self.kernel = self.kernel / K.sqrt(square_sum + self.eps) * self.wn_g
def sub_bias_initializer(_, *args, **kwagrs):
return K.concatenate([
self.sub_bias_initializer((self.sub_units, ), *args,
**kwagrs),
initializers.Ones()((self.sub_units, ), *args,
**kwagrs),
self.sub_bias_initializer((self.sub_units * 2, ),
*args, **kwagrs),
])
def build(self, input_shape):
self.gamma = self.add_weight(name='gamma',
shape=input_shape[-1:],
initializer=initializers.Ones(),
trainable=True)
self.beta = self.add_weight(name='beta',
shape=input_shape[-1:],
initializer=initializers.Zeros(),
trainable=True)
super(LayerNormalization, self).build(input_shape)
def bias_initializer(_, *args, **kwargs):
# Forget gate is the second slice.
init = K.concatenate([
self._bias_initializer(
(1, self.num_units), *args, **kwargs),
initializers.Ones()((1, self.num_units), *args, **kwargs),
self._bias_initializer(
(2, self.num_units), *args, **kwargs),
],
axis=0)
return array_ops.reshape(init, self.canonical_bias_shapes)
def build(self, input_shape, key_size, last_dim):
dtype = dtypes.as_dtype(self.dtype or K.floatx())
if not (dtype.is_floating or dtype.is_complex):
raise TypeError('Unable to build `PolymorphicDenseBase` layer with non-floating point '
'dtype %s' % (dtype,))
input_shape = tensor_shape.TensorShape(input_shape)
if tensor_shape.dimension_value(input_shape[-1]) is None:
raise ValueError('The last dimension of the inputs to `PolymorphicDenseBase` '
'should be defined. Found `None`.')
# Scalar that defines distance sensitivity when key is compared with
# keys map
self.similarity_sensitivity = self.add_weight(
'similarity_sensitivity',
shape=[self.modes],
initializer=initializers.Ones(),
regularizer=self.kernel_regularizer,
dtype=self.dtype,
trainable=True)
# Keys map that is compared with generated keys
self.keys_map = self.add_weight(
'keys_map',
shape=[self.modes, key_size],
initializer=self.kernel_initializer,
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
dtype=self.dtype,
trainable=True)
# Inputs processing weights map
self.kernels = self.add_weight(
'kernels',
shape=[self.modes, last_dim, self.units],
initializer=self.kernel_initializer,
regularizer=self.kernel_regularizer,
constraint=self.kernel_constraint,
dtype=self.dtype,
trainable=True)
if self.use_bias:
# Inputs processing biases map
self.biases = self.add_weight(
'biases',
shape=[self.modes, self.units],
initializer=self.bias_initializer,
regularizer=self.bias_regularizer,
constraint=self.bias_constraint,
dtype=self.dtype,
trainable=True)
else:
self.biases = None
super(PolymorphicDenseBase, self).build(input_shape)
def __init__(self,
weights=None,
axis=-1,
beta_init='zero',
gamma_init='one',
momentum=0.9,
**kwargs):
print("--Scale--init", axis)
# 参数**kwargs代表按字典方式继承父类
self.momentum = momentum
self.axis = axis
self.beta_init = initializers.Zeros()
self.gamma_init = initializers.Ones()
self.initial_weights = weights
super(Scale, self).__init__(**kwargs)