def call(self, X, mask=None):
input_shape = self.input_spec[0].shape
x = K.reshape(X[0], (-1, input_shape[2]))
target = X[1].flatten() if self.trainable else None
Y = h_softmax(x, K.shape(x)[0], self.output_dim,
self.n_classes, self.n_outputs_per_class,
self.W1, self.b1, self.W2, self.b2, target)
output_dim = 1 if self.trainable else self.output_dim
input_length = K.shape(X[0])[1]
y = K.reshape(Y, (-1, input_length, output_dim))
return y
def get_output(self, train=False):
X = self.get_input(train)
x = K.reshape(X[0], (-1, self.input_dim)) # (samples * timesteps, input_dim)
target = X[1].flatten() if train else None
Y = h_softmax(x, K.shape(x)[0], self.output_dim,
self.n_classes, self.n_outputs_per_class,
self.W1, self.b1, self.W2, self.b2, target)
flex_output = 1 if train else self.output_dim
output = K.reshape(Y, (-1, self.input_length, flex_output))
return output
# Second level of h_softmax
W2 = np.asarray(np.random.normal(0, 0.001,
size=(h_softmax_level1_size, d, h_softmax_level2_size)),
dtype=floatX)
W2 = shared(W2)
b2 = shared(
np.asarray(np.zeros((h_softmax_level1_size,
h_softmax_level2_size)), dtype=floatX))
#############
# Build graph
#############
# This only computes the output corresponding to the target
y_hat_tg = h_softmax(hid, m, output_size, h_softmax_level1_size,
h_softmax_level2_size, W1, b1, W2, b2, y_indexes)
# This computes all the outputs
# output = h_softmax(hid, m, output_size, h_softmax_level1_size,
# h_softmax_level2_size, W1, b1, W2, b2)
loss = -tensor.mean(tensor.log(y_hat_tg))
params.extend([W1, b1, W2, b2])
else:
raise ValueError
######################
# TRAINING FUNCTIONS #
######################
W2 = np.asarray(np.random.normal(0,
0.001,
size=(h_softmax_level1_size, d,
h_softmax_level2_size)),
dtype=floatX)
W2 = shared(W2)
b2 = shared(
np.asarray(np.zeros((h_softmax_level1_size, h_softmax_level2_size)),
dtype=floatX))
#############
# Build graph
#############
# This only computes the output corresponding to the target
y_hat_tg = h_softmax(hid, m, output_size, h_softmax_level1_size,
h_softmax_level2_size, W1, b1, W2, b2, y_indexes)
# This computes all the outputs
# output = h_softmax(hid, m, output_size, h_softmax_level1_size,
# h_softmax_level2_size, W1, b1, W2, b2)
loss = -tensor.mean(tensor.log(y_hat_tg))
params.extend([W1, b1, W2, b2])
else:
raise ValueError
######################
# TRAINING FUNCTIONS #
######################
grad = theano.grad(loss, [sub] + params)
def perform(self, x, y=None):
if y is not None:
return NN.h_softmax(x, x.shape[0], self.out_dim, self.h_level1_size, self.h_level2_size, self.params[0], self.params[1], self.params[2], self.params[3], y)
else:
return NN.h_softmax(x, x.shape[0], self.out_dim, self.h_level1_size, self.h_level2_size, self.params[0], self.params[1], self.params[2], self.params[3])
