def build(self):
if self.stateful:
self.reset_states()
else:
self.states = [None, None]
self.W_s = self.init((self.output_dim, self.output_dim))
self.W_t = self.init((self.output_dim, self.output_dim))
self.W_a = self.init((self.output_dim, self.output_dim))
self.w_e = K.zeros((self.output_dim,))
self.W_i = self.init((2 * self.output_dim, self.output_dim))
self.U_i = self.inner_init((self.output_dim, self.output_dim))
self.b_i = K.zeros((self.output_dim,))
self.W_f = self.init((2 * self.output_dim, self.output_dim))
self.U_f = self.inner_init((self.output_dim, self.output_dim))
self.b_f = self.forget_bias_init((self.output_dim,))
self.W_c = self.init((2 * self.output_dim, self.output_dim))
self.U_c = self.inner_init((self.output_dim, self.output_dim))
self.b_c = K.zeros((self.output_dim,))
self.W_o = self.init((2 * self.output_dim, self.output_dim))
self.U_o = self.inner_init((self.output_dim, self.output_dim))
self.b_o = K.zeros((self.output_dim,))
self.trainable_weights = [self.W_s, self.W_t, self.W_a, self.w_e,
self.W_i, self.U_i, self.b_i,
self.W_c, self.U_c, self.b_c,
self.W_f, self.U_f, self.b_f,
self.W_o, self.U_o, self.b_o]
def reset_states(self):
assert self.stateful, 'Layer must be stateful.'
if hasattr(self, 'states'):
K.set_value(self.states[0], np.zeros((self.batch_size, self.output_dim)))
K.set_value(self.states[1], np.zeros((self.batch_size, self.output_dim)))
else:
self.states = [K.zeros((self.batch_size, self.output_dim)),
K.zeros((self.batch_size, self.output_dim))]
def get_initial_states(self, X):
# build an all-zero tensor of shape (samples, output_dim)
initial_state = K.zeros_like(X) # (samples, timesteps, input_dim)
initial_state = K.sum(initial_state, axis=1) # (samples, input_dim)
reducer = K.zeros((self.output_dim, self.output_dim))
initial_state = K.dot(initial_state, reducer) # (samples, output_dim)
initial_states = [initial_state for _ in range(len(self.states))]
return initial_states
def get_initial_states(self, x):
# build an all-zero tensor of shape (samples, output_dim)
initial_state = K.zeros_like(x) # (samples, timesteps, input_dim)
initial_state = K.sum(initial_state, axis=1) # (samples, input_dim)
reducer = K.zeros((self.input_dim, self.output_dim))
initial_state = K.dot(initial_state, reducer) # (samples, output_dim)
initial_states = [initial_state for _ in range(len(self.states))]
if self.feed_layer is not None:
initial_states[1] = self.feed_layer
return initial_states
def get_initial_states(self, x):
# build an all-zero tensor of shape (samples, output_dim)
initial_state = K.zeros_like(x) # (samples, timesteps, input_dim)
initial_state = K.sum(initial_state, axis=1) # (samples, input_dim)
reducer = K.zeros((self.input_dim, self.output_dim))
initial_state = K.dot(initial_state, reducer) # (samples, output_dim)
initial_states = [initial_state for _ in range(len(self.states))]
if self.feed_layer is not None:
initial_states[1] = self.feed_layer
return initial_states
def build(self, input_shape):
self.input_spec = [InputSpec(shape=shape) for shape in input_shape]
input_dim = input_shape[1][2]
self.input_dim = input_dim
if self.stateful:
self.reset_states()
else:
self.states = [None, None]
self.W_s = self.init((self.output_dim, self.output_dim))
self.W_t = self.init((self.output_dim, self.output_dim))
self.W_a = self.init((self.output_dim, self.output_dim))
self.w_e = K.zeros((self.output_dim,))
self.W_i = self.init((2 * self.output_dim, self.output_dim))
self.U_i = self.inner_init((self.output_dim, self.output_dim))
self.b_i = K.zeros((self.output_dim,))
self.W_f = self.init((2 * self.output_dim, self.output_dim))
self.U_f = self.inner_init((self.output_dim, self.output_dim))
self.b_f = self.forget_bias_init((self.output_dim,))
self.W_c = self.init((2 * self.output_dim, self.output_dim))
self.U_c = self.inner_init((self.output_dim, self.output_dim))
self.b_c = K.zeros((self.output_dim,))
self.W_o = self.init((2 * self.output_dim, self.output_dim))
self.U_o = self.inner_init((self.output_dim, self.output_dim))
self.b_o = K.zeros((self.output_dim,))
self.trainable_weights = [self.W_s, self.W_t, self.W_a, self.w_e,
self.W_i, self.U_i, self.b_i,
self.W_c, self.U_c, self.b_c,
self.W_f, self.U_f, self.b_f,
self.W_o, self.U_o, self.b_o]
def build(self, input_shape):
self.input_spec = [InputSpec(shape=shape) for shape in input_shape]
input_dim = input_shape[1][2]
self.input_dim = input_dim
if self.stateful:
self.reset_states()
else:
self.states = [None, None]
self.W_s = self.init((self.output_dim, self.output_dim))
self.W_t = self.init((self.output_dim, self.output_dim))
self.W_a = self.init((self.output_dim, self.output_dim))
self.w_e = K.zeros((self.output_dim, ))
self.W_i = self.init((2 * self.output_dim, self.output_dim))
self.U_i = self.inner_init((self.output_dim, self.output_dim))
self.b_i = K.zeros((self.output_dim, ))
self.W_f = self.init((2 * self.output_dim, self.output_dim))
self.U_f = self.inner_init((self.output_dim, self.output_dim))
self.b_f = self.forget_bias_init((self.output_dim, ))
self.W_c = self.init((2 * self.output_dim, self.output_dim))
self.U_c = self.inner_init((self.output_dim, self.output_dim))
self.b_c = K.zeros((self.output_dim, ))
self.W_o = self.init((2 * self.output_dim, self.output_dim))
self.U_o = self.inner_init((self.output_dim, self.output_dim))
self.b_o = K.zeros((self.output_dim, ))
self.trainable_weights = [
self.W_s, self.W_t, self.W_a, self.w_e, self.W_i, self.U_i,
self.b_i, self.W_c, self.U_c, self.b_c, self.W_f, self.U_f,
self.b_f, self.W_o, self.U_o, self.b_o
]
