def __init__(self, input_dim, output_dim, **kwargs): """Initialization. Weights are randomly initialized from a normal distribution. Biases are initialized to zero. Args: input_dim: input dimensionality to layer output_dim: output dimensionality from layer mu (kwargs): set the mean for random initialization sigma (kwargs): set the standard deviation for random initialization """ Layer.__init__(self) self.input_dim = input_dim self.output_dim = output_dim self.mu = kwargs["mu"] if "mu" in kwargs else LinearLayer.DEFAULT_INITIAL_MU self.sigma = kwargs["sigma"] if "sigma" in kwargs else LinearLayer.DEFAULT_INITIAL_SIGMA # initialize weights if output_dim == 1: shape = (input_dim,) else: shape = (output_dim, input_dim) self.W = np.random.normal(self.mu, self.sigma, shape) # initialize biases if output_dim == 1: self.b = 0 else: self.b = np.zeros(output_dim)
def __init__(self, input_dim, output_dim, **kwargs):
"""Initialization.
Weights are randomly initialized from a normal distribution.
Biases are initialized to zero.
Args:
input_dim: input dimensionality to layer
output_dim: output dimensionality from layer
mu (kwargs): set the mean for random initialization
sigma (kwargs): set the standard deviation for random initialization
"""
Layer.__init__(self)
self.input_dim = input_dim
self.output_dim = output_dim
self.mu = kwargs[
"mu"] if "mu" in kwargs else LinearLayer.DEFAULT_INITIAL_MU
self.sigma = kwargs[
"sigma"] if "sigma" in kwargs else LinearLayer.DEFAULT_INITIAL_SIGMA
# initialize weights
if output_dim == 1:
shape = (input_dim, )
else:
shape = (output_dim, input_dim)
self.W = np.random.normal(self.mu, self.sigma, shape)
# initialize biases
if output_dim == 1:
self.b = 0
else:
self.b = np.zeros(output_dim)
def __init__(self, input_dim, output_dim, **kwargs):
Layer.__init__(self)
self.input_dim = input_dim
self.output_dim = output_dim
self.mu = kwargs[
"mu"] if "mu" in kwargs else LinearLayer.DEFAULT_INITIAL_MU
self.sigma = kwargs[
"sigma"] if "sigma" in kwargs else LinearLayer.DEFAULT_INITIAL_SIGMA
if output_dim == 1:
shape = (input_dim, )
else:
shape = (output_dim, input_dim)
self.W = np.random.normal(self.mu, self.sigma, shape)
if output_dim == 1:
self.b = 0
else:
self.b = np.zeros(output_dim)
def __init__(self):
Layer.__init__(self)
def __init__(self): Layer.__init__(self)
