上一篇介绍了如何编写单层的LSTM网络。对于一些复杂的序列，需要用到多层的网络进行学习。这里介绍如何利用TensorFlow（r1.1）编写多层LSTM网络。
建立模型
首先利用tf.contrib.rnn.MultiRNNCell将多个BasicLSTMCell单元汇总为一个。值得注意的是，每次添加一个单元需要重新调用一次BasicLSTMCell。因为该函数每次都会声明一次内部变量，如果不这么做则会reuse这些变量，从而产生错误。
# Forward passes
cells = []
for n in range(num_layers):
cells.append(tf.contrib.rnn.BasicLSTMCell(state_size, state_is_tuple=True))
cell = tf.contrib.rnn.MultiRNNCell(cells, state_is_tuple=True)
为每一层的初始状态设置初始值。也可以采用.zero_state方法生成初始值，但是这样就不能对中间状态进行显示控制。具体根据实际应用选择。
init_state = tf.placeholder(tf.float32, [num_layers, 2, batch_size, state_size])
state_per_layer_list = tf.stack(init_state, axis=0)
rnn_tuple_state = tuple(
[tf.contrib.rnn.LSTMStateTuple(state_per_layer_list[idx][0], state_per_layer_list[idx][1]) for idx in range(num_layers)]
)
# init_state = cell.zero_state(batch_size, tf.float32)
损失函数
基于最后一层网络的输出状态进行预测估计。
logits_series = []
for state in states_series:
logits_series.append(tf.matmul(state[-1][0], W1) + tf.matmul(state[-1][1], W2) + b2)
predictions_series = [tf.nn.softmax(logits) for logits in logits_series]
模型训练
利用numpy计算训练中的初始值。
_current_state = np.zeros((num_layers, 2, batch_size, state_size))
全部代码
from __future__ import print_function, division
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
num_epochs = 100
total_series_length = 50000
truncated_backprop_length = 15
state_size = 4
num_classes = 2
echo_step = 3
batch_size = 5
num_batches = total_series_length//batch_size//truncated_backprop_length
num_layers = 3
def generateData():
x = np.array(np.random.choice(2, total_series_length, p=[0.5, 0.5]))
y = np.roll(x, echo_step)
y[0:echo_step] = 0
x = x.reshape((batch_size, -1)) # The first index changing slowest, subseries as rows
y = y.reshape((batch_size, -1))
return (x, y)
batchX_placeholder = tf.placeholder(tf.float32, [batch_size, truncated_backprop_length])
batchY_placeholder = tf.placeholder(tf.int32, [batch_size, truncated_backprop_length])
# Unpack columns
inputs_series = tf.split(batchX_placeholder, truncated_backprop_length, axis=1)
labels_series = tf.unstack(batchY_placeholder, axis=1)
# Forward passes
cell = tf.contrib.rnn.BasicLSTMCell(state_size, state_is_tuple=True)
cells = []
for n in range(num_layers):
cells.append(tf.contrib.rnn.BasicLSTMCell(state_size, state_is_tuple=True))
stacked_lstm = tf.contrib.rnn.MultiRNNCell(cells, state_is_tuple=True)
init_state = tf.placeholder(tf.float32, [num_layers, 2, batch_size, state_size])
state_per_layer_list = tf.stack(init_state, axis=0)
rnn_tuple_state = tuple(
[tf.contrib.rnn.LSTMStateTuple(state_per_layer_list[idx][0], state_per_layer_list[idx][1]) for idx in range(num_layers)]
)
# init_state = stacked_lstm.zero_state(batch_size, tf.float32)
current_state = rnn_tuple_state
states_series = []
for current_input in inputs_series:
with tf.variable_scope('rnn') as vs:
try:
output, current_state = stacked_lstm(current_input, current_state)
except:
vs.reuse_variables()
output, current_state = stacked_lstm(current_input, current_state)
states_series.append(current_state)
W1 = tf.Variable(np.random.rand(state_size, num_classes),dtype=tf.float32)
b1 = tf.Variable(np.zeros((1,num_classes)), dtype=tf.float32)
W2 = tf.Variable(np.random.rand(state_size, num_classes),dtype=tf.float32)
b2 = tf.Variable(np.zeros((1,num_classes)), dtype=tf.float32)
logits_series = []
for state in states_series:
logits_series.append(tf.matmul(state[-1][0], W1) + tf.matmul(state[-1][1], W2) + b2)
predictions_series = [tf.nn.softmax(logits) for logits in logits_series]
losses = [tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logits, labels=labels) for logits, labels in zip(logits_series,labels_series)]
total_loss = tf.reduce_mean(losses)
train_step = tf.train.AdagradOptimizer(0.3).minimize(total_loss)
def plot(loss_list, predictions_series, batchX, batchY):
plt.subplot(2, 3, 1)
plt.cla()
plt.plot(loss_list)
for batch_series_idx in range(5):
one_hot_output_series = np.array(predictions_series)[:, batch_series_idx, :]
single_output_series = np.array([(1 if out[0] < 0.5 else 0) for out in one_hot_output_series])
plt.subplot(2, 3, batch_series_idx + 2)
plt.cla()
plt.axis([0, truncated_backprop_length, 0, 2])
left_offset = range(truncated_backprop_length)
plt.bar(left_offset, batchX[batch_series_idx, :], width=1, color="blue")
plt.bar(left_offset, batchY[batch_series_idx, :] * 0.5, width=1, color="red")
plt.bar(left_offset, single_output_series * 0.3, width=1, color="green")
plt.draw()
plt.pause(0.0001)
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
plt.ion()
plt.figure()
plt.show()
loss_list = []
for epoch_idx in range(num_epochs):
x,y = generateData()
_current_state = np.zeros((num_layers, 2, batch_size, state_size))
print("New data, epoch", epoch_idx)
for batch_idx in range(num_batches):
start_idx = batch_idx * truncated_backprop_length
end_idx = start_idx + truncated_backprop_length
batchX = x[:,start_idx:end_idx]
batchY = y[:,start_idx:end_idx]
_total_loss, _train_step, _current_state, _predictions_series = sess.run(
[total_loss, train_step, current_state, predictions_series],
feed_dict={
batchX_placeholder: batchX,
batchY_placeholder: batchY,
init_state: _current_state
})
loss_list.append(_total_loss)
if batch_idx%100 == 0:
print("Step",batch_idx, "Batch loss", _total_loss)
plot(loss_list, _predictions_series, batchX, batchY)
plt.ioff()
plt.show()
参考文献：
