除此之外, 这篇博客: 基于CNN-LSTM的手写数字识别与应用实现（附tensorflow代码讲解）中的 CNN-LSTM模型代码实现 部分也许能够解决你的问题, 你可以仔细阅读以下内容或跳转源博客中阅读:
CNN-LSTM的tensorflow版本实现：
def cnn_lstm(x):
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
def bias_variable(shape):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial)
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
def max_pool_2x2(x):
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME')
with tf.variable_scope('input'):
x = tf.reshape(x, [-1, 28, 28, 1])
with tf.variable_scope('conv_pool_1'):
W_conv1 = weight_variable([5, 5, 1, 32])
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)
with tf.variable_scope('conv_pool_2'):
W_conv2 = weight_variable([5, 5, 32, 64])
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
h_pool2 = max_pool_2x2(h_conv2)
X_in = tf.reshape(h_pool2, [-1, 49, 64])
X_in = tf.transpose(X_in, [0, 2, 1])
with tf.variable_scope('lstm'):
lstm_cell = tf.contrib.rnn.BasicLSTMCell(
128, forget_bias=1.0, state_is_tuple=True)
outputs, states = tf.nn.dynamic_rnn(
lstm_cell, X_in, time_major=False, dtype=tf.float32)
W_lstm = weight_variable([128, 10])
b_lstm = bias_variable([10])
outputs = tf.unstack(tf.transpose(outputs, [1, 0, 2]))
y = tf.nn.softmax(tf.matmul(outputs[-1], W_lstm) + b_lstm)
train_vars = tf.trainable_variables()
return y, train_vars
上面已经定义好了CNN-LSTM整个完整的模型，下面我们将在另一个.py文件下调用他。
from model import cnn_lstm
接着开始调用（下载）MNIST数据集，其中one_hot=True，该参数的功能主要是将图片向量转换成one_hot类型的张量输出。
data = input_data.read_data_sets('MNIST_data', one_hot=True)
在调用CNN-LSTM模型后，需要在训练模型的这个.py文件定义模型，其中x = tf.placeholder为输入变量占位符，在训练前就要指定。
with tf.variable_scope('cnn_lstm'):
x = tf.placeholder(tf.float32, [None, 784], name='x')
y, variables = cnn_lstm(x)
接着为了训练模型，需要首先进行添加一个新的占位符用于输入正确值，接着定义交叉熵损失函数、学习速率等。
y_ = tf.placeholder('float', [None, 10])
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=y, labels=y_))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
correct_pred = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
开始训练模型之前，在Session里面启动模型，其中accuracy.eval(feed_dict={x: batch[0], y_: batch[1]})计算所学习到的模型的正确率。
with tf.Session() as sess:
merged_summary_op = tf.summary.merge_all()
summary_write = tf.summary.FileWriter('tmp/mnist_log/1', sess.graph)
summary_write.add_graph(sess.graph)
sess.run(tf.global_variables_initializer())
for i in range(20000):
batch = data.train.next_batch(50)
if i % 100 == 0:
train_accuracy = accuracy.eval(feed_dict={x: batch[0], y_: batch[1]})
print("Step %d, training accuracy %g" % (i, train_accuracy))
sess.run(train_step, feed_dict={x: batch[0], y_: batch[1]})
result = []
for i in range(2000):
batch = data.test.next_batch(50)
result.append(sess.run(accuracy, feed_dict={x: batch[0], y_: batch[1]}))
print(sum(result)/len(result))
训练代码完整如下：
data = input_data.read_data_sets('MNIST_data', one_hot=True)
with tf.variable_scope('cnn_lstm'):
x = tf.placeholder(tf.float32, [None, 784], name='x')
y, variables = cnn_lstm(x)
y_ = tf.placeholder('float', [None, 10])
cross_entropy = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(logits=y, labels=y_))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
correct_pred = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
saver = tf.train.Saver(variables)
with tf.Session() as sess:
merged_summary_op = tf.summary.merge_all()
summary_write = tf.summary.FileWriter('tmp/mnist_log/1', sess.graph)
summary_write.add_graph(sess.graph)
sess.run(tf.global_variables_initializer())
for i in range(20000):
batch = data.train.next_batch(50)
if i % 100 == 0:
train_accuracy = accuracy.eval(feed_dict={x: batch[0], y_: batch[1]})
print("Step %d, training accuracy %g" % (i, train_accuracy))
sess.run(train_step, feed_dict={x: batch[0], y_: batch[1]})
result = []
for i in range(2000):
batch = data.test.next_batch(50)
result.append(sess.run(accuracy, feed_dict={x: batch[0], y_: batch[1]}))
print(sum(result)/len(result))
训练结束后，将CNN-LSTM模型的训练参数进行保存，其实现代码如下：
path = saver.save(sess,os.path.join(os.path.dirname(__file__), 'data', 'cnn_lstm.ckpt'), write_meta_graph=False,write_state=False )
