def __init__(self, input_dim, output_dim, init='glorot_uniform', activation='tanh', name='Dense',
learn_bias=True, negative_bias=False):
super(Dense, self).__init__()
self.init = initializations.get(init)
self.activation = activations.get(activation)
self.input_dim = input_dim
self.output_dim = output_dim
self.linear = (activation == 'linear')
# self.input = T.matrix()
self.W = self.init((self.input_dim, self.output_dim))
if not negative_bias:
self.b = shared_zeros((self.output_dim))
else:
self.b = shared_ones((self.output_dim))
self.learn_bias = learn_bias
if self.learn_bias:
self.params = [self.W, self.b]
else:
self.params = [self.W]
if name is not None:
self.set_name(name)
def __init__(self,
input_dim1,
input_dim2,
output_dim,
init='glorot_uniform',
activation='tanh',
name='Dense',
learn_bias=True):
super(Dense2, self).__init__()
self.init = initializations.get(init)
self.activation = activations.get(activation)
self.input_dim1 = input_dim1
self.input_dim2 = input_dim2
self.output_dim = output_dim
self.linear = (activation == 'linear')
# self.input = T.matrix()
self.W1 = self.init((self.input_dim1, self.output_dim))
self.W2 = self.init((self.input_dim2, self.output_dim))
self.b = shared_zeros((self.output_dim))
self.learn_bias = learn_bias
if self.learn_bias:
self.params = [self.W1, self.W2, self.b]
else:
self.params = [self.W1, self.W2]
if name is not None:
self.set_name(name)
def __init__(self, input_dim, output_dim, init='uniform', name=None):
super(Embedding, self).__init__()
self.init = initializations.get(init)
self.input_dim = input_dim
self.output_dim = output_dim
self.W = self.init((self.input_dim, self.output_dim))
self.params = [self.W]
if name is not None:
self.set_name(name)
def __init__(self, input_dim, output_dim, init=None, activation='tanh', name='Bias'):
super(Constant, self).__init__()
assert input_dim == output_dim, 'Bias Layer needs to have the same input/output nodes.'
self.init = initializations.get(init)
self.activation = activations.get(activation)
self.input_dim = input_dim
self.output_dim = output_dim
self.b = shared_zeros(self.output_dim)
self.params = [self.b]
if name is not None:
self.set_name(name)
def __init__(self, input_dim, input_wdth, init='glorot_uniform',
activation='tanh', name='Bias', has_input=True):
super(MemoryLinear, self).__init__()
self.init = initializations.get(init)
self.activation = activations.get(activation)
self.input_dim = input_dim
self.input_wdth = input_wdth
self.b = self.init((self.input_dim, self.input_wdth))
self.params = [self.b]
if has_input:
self.P = self.init((self.input_dim, self.input_wdth))
self.params += [self.P]
if name is not None:
self.set_name(name)