def _get_bias_init(config):
config = config or {
'typename': 'ConstantInitializer',
'args': {
'value': 0.1
}
}
return getter.get_initializer(config['typename'])(**config.get('args', {}))
def _get_bias_init(config):
"""Make bias initializer. Default to Constant (0.1)"""
config = config or {
'typename': 'ConstantInitializer',
'args': {
'value': 0.1
}
}
return get_initializer(config['typename'])(**config.get('args', {}))
def _instantiate_parameters(self, input_shape, dtype):
dim = len(input_shape)
shape = tuple(input_shape[i] for i in range(dim) if i == 1)
self._axes = tuple(i for i in range(dim) if not i == 1)
self._pattern = tuple((0 if i == 1 else 'x') for i in range(dim))
_LG.debug(' Shape: %s', shape)
_LG.debug(' Axes: %s', self._axes)
_LG.debug(' Pattern: %s', self._pattern)
const_init = get_initializer('ConstantInitializer')
if self._parameter_variables['mean'] is None:
mean = wrapper.get_variable(name='mean',
shape=shape,
trainable=False,
initializer=const_init(0),
dtype=dtype)
self.set_parameter_variables(mean=mean)
if self._parameter_variables['var'] is None:
var = wrapper.get_variable(name='var',
shape=shape,
trainable=False,
initializer=const_init(1),
dtype=dtype)
self.set_parameter_variables(var=var)
if self._parameter_variables['scale'] is None:
scale_val = self.args['scale']
scale = wrapper.get_variable(name='scale',
shape=shape,
trainable=True,
initializer=const_init(scale_val),
dtype=dtype)
self.set_parameter_variables(scale=scale)
if self._parameter_variables['offset'] is None:
offset_val = self.args['offset']
offset = wrapper.get_variable(name='offset',
shape=shape,
trainable=True,
initializer=const_init(offset_val),
dtype=dtype)
self.set_parameter_variables(offset=offset)
def _instantiate_parameters(self, input_shape):
dim, fmt = len(input_shape), luchador.get_nn_conv_format()
channel = 1 if dim == 2 or fmt == 'NCHW' else 3
self._axes = tuple(i for i in range(dim) if not i == channel)
shape = tuple(input_shape[i] for i in range(dim) if i == channel)
const_init = get_initializer('ConstantInitializer')
if self.get_parameter_variable('mean') is None:
mean = wrapper.get_variable(name='mean',
shape=shape,
initializer=const_init(0),
trainable=False)
self.set_parameter_variables(mean=mean)
if self.get_parameter_variable('var') is None:
var = wrapper.get_variable(name='var',
shape=shape,
initializer=const_init(1),
trainable=False)
self.set_parameter_variables(var=var)
if self.get_parameter_variable('scale') is None:
scale = wrapper.get_variable(name='scale',
shape=shape,
trainable=True,
initializer=const_init(
self.args['scale']))
self.set_parameter_variables(scale=scale)
if self.get_parameter_variable('offset') is None:
offset = wrapper.get_variable(name='offset',
shape=shape,
trainable=True,
initializer=const_init(
self.args['offset']))
self.set_parameter_variables(offset=offset)
def _get_weight_init(config):
config = config or {'typename': 'XavierInitializer'}
return getter.get_initializer(config['typename'])(**config.get('args', {}))
def _get_filter_init(config):
"""Make filter initializer. Default to Xavier"""
config = config or {'typename': 'XavierInitializer'}
return get_initializer(config['typename'])(**config.get('args', {}))