def __call__(self, inputs, state, scope = None):
with tf.variable_scope(scope or type(self).__name__):
with tf.variable_scope("Gates"):
reset, update = tf.split(
1,
2,
linear(
[inputs, states],
2 * self._num_units,
bias = True,
bias_start = 1.0
)
)
reset, update = tf.sigmoid(reset), tf.sigmoid(update)
with tf.variable_scope("Candidate"):
candidate = linear(
[inputs, reset * state],
self._num_units,
bias = True
)
candidate = tf.tanh(candidate)
new_state = update * state + (1 - update) * candidate
return new_state, new_state
def __call__(self, inputs, state, scope = None):
with tf.variable_scope(scope or type(self).__name__):
c, h = tf.split(1, 2, state)
concat = linear(
[inputs, h],
4 * self._num_units,
bias = True
)
i, j, f, o = tf.split(1, 4, concat)
new_c = c * tf.sigmoid(f + self._forget_bias) + tf.sigmoid(i) * tf.tanh(j)
new_h = tf.tanh(new_c) * tf.sigmoid(o)
new_state = tf.concat(1, [new_c, new_h])
return new_h, new_state
def __call__(self, inputs, state, scope = None):
with tf.variable_scope(scope or type(self).__name__):
output = tf.tanh(linear([inputs, state], self._num_units, bias = True))
return output, output