def __init__(self, hidden_neurons=100, activation_func='tanh', *args,
**kwargs):
super(SLmNce, self).__init__(*args, **kwargs)
self.name = 'slm'
self.activation_func = activation_func
self.hidden_neurons = hidden_neurons
self.updatable_parameters.extend(['W1', 'W2'])
rand_values = random_value_GloBen10(
(self.context_size * self.emb_size + 1, self.hidden_neurons),
floatX, np.random.RandomState(7816))
self.W1 = theano.shared(value=rand_values, name='W')
rand_values = random_value_GloBen10(
(self.hidden_neurons + 1, self.emb_size),
floatX, np.random.RandomState(7817))
self.W2 = theano.shared(value=rand_values, name='W')
def __init__(self, *args, **kwargs):
super(VLblNce, self).__init__(*args, **kwargs)
self.regularize.append('C')
rand_values = random_value_GloBen10((self.context_size, self.emb_size),
floatX, np.random.RandomState(2341),
(self.context_size * self.emb_size, self.emb_size))
self.C = theano.shared(value=rand_values, name='C')
def __init__(self, *args, **kwargs):
super(LblNce, self).__init__(*args, **kwargs)
self.name = 'lbl'
self.updatable_parameters.append('W')
self.regularize.append('W')
rand_values = random_value_GloBen10(
(self.context_size * self.emb_size, self.emb_size),
floatX, np.random.RandomState(7816))
self.W = theano.shared(value=rand_values, name='W')
def __init__(self,
hidden_neurons=100,
activation_func='tanh',
*args,
**kwargs):
super(SLmNce, self).__init__(*args, **kwargs)
self.name = 'slm'
self.activation_func = activation_func
self.hidden_neurons = hidden_neurons
self.updatable_parameters.extend(['W1', 'W2'])
rand_values = random_value_GloBen10(
(self.context_size * self.emb_size + 1, self.hidden_neurons),
floatX, np.random.RandomState(7816))
self.W1 = theano.shared(value=rand_values, name='W')
rand_values = random_value_GloBen10(
(self.hidden_neurons + 1, self.emb_size), floatX,
np.random.RandomState(7817))
self.W2 = theano.shared(value=rand_values, name='W')
def __init__(self, *args, **kwargs):
super(LblNce, self).__init__(*args, **kwargs)
self.name = 'lbl'
self.updatable_parameters.append('W')
self.regularize.append('W')
rand_values = random_value_GloBen10(
(self.context_size * self.emb_size, self.emb_size), floatX,
np.random.RandomState(7816))
self.W = theano.shared(value=rand_values, name='W')
def __init__(self, *args, **kwargs):
super(VLblNce, self).__init__(*args, **kwargs)
self.regularize.append('C')
rand_values = random_value_GloBen10(
(self.context_size, self.emb_size), floatX,
np.random.RandomState(2341),
(self.context_size * self.emb_size, self.emb_size))
self.C = theano.shared(value=rand_values, name='C')
def __init__(self,
name='HiddenLayer',
shape=(0, 0),
w_values=None,
activation=T.tanh):
super(HiddenLayer, self).__init__(name)
self.activation = activation
if w_values is None:
input_dim, output_dim, = shape
w_values = random_value_GloBen10((input_dim, output_dim), floatX)
if activation == theano.tensor.nnet.sigmoid:
w_values *= 4
self.fan_in = w_values.shape[0]
self.clipping = False
self.threshold = 1e12
self.weights = theano.shared(value=w_values,
name='weights_' + self.name)