Create a gist now

Instantly share code, notes, and snippets.

What would you like to do?
An implementation of multiplicative LSTM in TensorFlow
# Copyright (C) 2017 by Akira TAMAMORI
# This program is free software; you can redistribute it and/or modify it under
# the terms of the GNU General Public License as published by the Free Software
# Foundation, either version 3 of the License, or (at your option) any later
# version.
# This program is distributed in the hope that it will be useful, but WITHOUT
# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
# FOR A PARTICULAR PURPOSE. See the GNU General Public License for more
# details.
# You should have received a copy of the GNU General Public License along with
# this program. If not, see <>.
# Copyright 2015 The TensorFlow Authors. All Rights Reserved.
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# See the License for the specific language governing permissions and
# limitations under the License.
# Notice:
# This file is tested on TensorFlow v0.12.0 only.
import numpy as np
import tensorflow as tf
from tensorflow.python.ops.rnn_cell import RNNCell
# Thanks to '' of Project RNN Enhancement:
def orthogonal_initializer(scale=1.0):
def _initializer(shape, dtype=tf.float32):
flat_shape = (shape[0],[1:]))
a = np.random.normal(0.0, 1.0, flat_shape)
u, _, v = np.linalg.svd(a, full_matrices=False)
q = u if u.shape == flat_shape else v
q = q.reshape(shape)
return tf.constant(scale * q[:shape[0], :shape[1]], dtype=tf.float32)
return _initializer
class MultiplicativeLSTMCell(RNNCell):
"""Multiplicative LSTM.
Ben Krause, Liang Lu, Iain Murray, and Steve Renals,
"Multiplicative LSTM for sequence modelling, "
in Workshop Track of ICLA 2017,
def __init__(self, num_units,
"""Initialize the parameters for an LSTM cell.
num_units: int, The number of units in the LSTM cell.
use_peepholes: bool, set True to enable diagonal/peephole
cell_clip: (optional) A float value, if provided the cell state
is clipped by this value prior to the cell output activation.
initializer: (optional) The initializer to use for the weight
num_proj: (optional) int, The output dimensionality for
the projection matrices. If None, no projection is performed.
forget_bias: Biases of the forget gate are initialized by default
to 1 in order to reduce the scale of forgetting at the beginning of
the training.
activation: Activation function of the inner states.
self.num_units = num_units
self.use_peepholes = use_peepholes
self.cell_clip = cell_clip
self.num_proj = num_proj
self.proj_clip = proj_clip
self.initializer = initializer
self.forget_bias = forget_bias
self.state_is_tuple = state_is_tuple
self.activation = activation
if num_proj:
self._state_size = (
tf.nn.rnn_cell.LSTMStateTuple(num_units, num_proj)
if state_is_tuple else num_units + num_proj)
self._output_size = num_proj
self._state_size = (
tf.nn.rnn_cell.LSTMStateTuple(num_units, num_units)
if state_is_tuple else 2 * num_units)
self._output_size = num_units
def state_size(self):
return self._state_size
def output_size(self):
return self._output_size
def __call__(self, inputs, state, scope=None):
num_proj = self.num_units if self.num_proj is None else self.num_proj
if self.state_is_tuple:
(c_prev, h_prev) = state
c_prev = tf.slice(state, [0, 0], [-1, self.num_units])
h_prev = tf.slice(state, [0, self.num_units], [-1, num_proj])
dtype = inputs.dtype
input_size = inputs.get_shape().with_rank(2)[1]
with tf.variable_scope(scope or type(self).__name__):
if input_size.value is None:
raise ValueError(
"Could not infer input size from inputs.get_shape()[-1]")
with tf.variable_scope("Multipli_Weight"):
concat = _linear([inputs, h_prev], 2 * self.num_units, True)
Wx, Wh = tf.split(1, 2, concat)
m = Wx * Wh # equation (18)
with tf.variable_scope("LSTM_Weight"):
lstm_matrix = _linear([inputs, m], 4 * self.num_units, True)
i, j, f, o = tf.split(1, 4, lstm_matrix)
# Diagonal connections
if self.use_peepholes:
w_f_diag = tf.get_variable(
"W_F_diag", shape=[self.num_units], dtype=dtype)
w_i_diag = tf.get_variable(
"W_I_diag", shape=[self.num_units], dtype=dtype)
w_o_diag = tf.get_variable(
"W_O_diag", shape=[self.num_units], dtype=dtype)
if self.use_peepholes:
c = c_prev * tf.sigmoid(f + self.forget_bias +
w_f_diag * c_prev) + \
tf.sigmoid(i + w_i_diag * c_prev) * j
c = c_prev * tf.sigmoid(f + self.forget_bias) + \
tf.sigmoid(i) * j
if self.cell_clip is not None:
c = tf.clip_by_value(c, -self.cell_clip, self.cell_clip)
if self.use_peepholes:
h = tf.sigmoid(o + w_o_diag * c) * \
self.activation(c * (o + w_o_diag * c))
h = self.activation(c * o)
if self.num_proj is not None:
w_proj = tf.get_variable(
"W_P", [self.num_units, num_proj], dtype=dtype)
h = tf.matmul(h, w_proj)
if self.proj_clip is not None:
h = tf.clip_by_value(h, -self.proj_clip, self.proj_clip)
new_state = (tf.nn.rnn_cell.LSTMStateTuple(c, h)
if self.state_is_tuple else tf.concat(1, [c, h]))
return h, new_state
def _linear(args, output_size, bias, bias_start=0.0, scope=None):
"""Linear map: sum_i(args[i] * W[i]), where W[i] is a variable.
args: a 2D Tensor or a list of 2D, batch x n, Tensors.
output_size: int, second dimension of W[i].
bias: boolean, whether to add a bias term or not.
bias_start: starting value to initialize the bias; 0 by default.
scope: VariableScope for the created subgraph; defaults to "Linear".
A 2D Tensor with shape [batch x output_size] equal to
sum_i(args[i] * W[i]), where W[i]s are newly created matrices.
ValueError: if some of the arguments has unspecified or wrong shape.
if args is None or (isinstance(args, (list, tuple)) and not args):
raise ValueError("`args` must be specified")
if not isinstance(args, (list, tuple)):
args = [args]
# Calculate the total size of arguments on dimension 1.
total_arg_size = 0
shapes = [a.get_shape().as_list() for a in args]
for shape in shapes:
if len(shape) != 2:
raise ValueError(
"Linear is expecting 2D arguments: %s" % str(shapes))
if not shape[1]:
raise ValueError(
"Linear expects shape[1] of arguments: %s" % str(shapes))
total_arg_size += shape[1]
# Now the computation.
with tf.variable_scope(scope or "Linear"):
matrix = tf.get_variable("Matrix", [total_arg_size, output_size])
if len(args) == 1:
res = tf.matmul(args[0], matrix)
res = tf.matmul(tf.concat(1, args), matrix)
if not bias:
return res
bias_term = tf.get_variable(
"Bias", [output_size],
return res + bias_term

This comment has been minimized.

Show comment Hide comment

lspinheiro Jun 22, 2017

If I may ask, did you test for performance? Did this implementation show the faster convergence reported in the paper?

If I may ask, did you test for performance? Did this implementation show the faster convergence reported in the paper?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment