一看就懂的Tensorflow实战（LSTM）

from __future__ import print_function

import tensorflow as tf
from tensorflow.contrib import rnn

# Import MNIST data
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("./data/", one_hot=True)

Extracting ./data/train-images-idx3-ubyte.gz
Extracting ./data/train-labels-idx1-ubyte.gz
Extracting ./data/t10k-images-idx3-ubyte.gz
Extracting ./data/t10k-labels-idx1-ubyte.gz

# 训练参数
learning_rate = 0.001 # 学习率
training_steps = 10000 # 总迭代次数
batch_size = 128 # 批量大小
display_step = 200

# 网络参数
num_input = 28 # MNIST数据集图片: 28*28
timesteps = 28 # timesteps
num_hidden = 128 # 隐藏层神经元数
num_classes = 10 # MNIST 数据集类别数 (0-9 digits)

# 定义输入
X = tf.placeholder("float", [None, timesteps, num_input])
Y = tf.placeholder("float", [None, num_classes])

# 定义权重和偏置
# weights矩阵[128, 10]
weights = {
    'out': tf.Variable(tf.random_normal([num_hidden, num_classes]))
}
biases = {
    'out': tf.Variable(tf.random_normal([num_classes]))
}

# 定义LSTM网络
def LSTM(x, weights, biases):

    # Prepare data shape to match `rnn` function requirements
    # 输入数据x的shape: (batch_size, timesteps, n_input)
    # 需要的shape: 按 timesteps 切片，得到 timesteps 个 (batch_size, n_input)

    # 对x进行切分
    # tf.unstack(value,num=None,axis=0,name='unstack')
    # value：要进行分割的tensor
    # axis：整数，打算进行切分的维度
    # num：整数，axis（打算切分）维度的长度
    x = tf.unstack(x, timesteps, 1)

    # 定义一个lstm cell，即上面图示LSTM中的A
    # n_hidden表示神经元的个数，forget_bias就是LSTM们的忘记系数，如果等于1，就是不会忘记任何信息。如果等于0，就都忘记。
    lstm_cell = rnn.BasicLSTMCell(num_hidden, forget_bias=1.0)

    # 得到 lstm cell 输出
    # 输出output和states
    # outputs是一个长度为T的列表，通过outputs[-1]取出最后的输出
    # state是最后的状态
    outputs, states = rnn.static_rnn(lstm_cell, x, dtype=tf.float32)

    # 线性激活
    # 矩阵乘法
    return tf.matmul(outputs[-1], weights['out']) + biases['out']


logits = LSTM(X, weights, biases)
prediction = tf.nn.softmax(logits)

# 定义损失函数和优化器
loss_op = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(
    logits=logits, labels=Y))
optimizer = tf.train.GradientDescentOptimizer(learning_rate=learning_rate)
train_op = optimizer.minimize(loss_op)

# 模型评估(with test logits, for dropout to be disabled)
correct_pred = tf.equal(tf.argmax(prediction, 1), tf.argmax(Y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))

# 初始化全局变量
init = tf.global_variables_initializer()

# Start training
with tf.Session() as sess:

    # Run the initializer
    sess.run(init)

    for step in range(1, training_steps+1):
        batch_x, batch_y = mnist.train.next_batch(batch_size)
        # Reshape data to get 28 seq of 28 elements
        batch_x = batch_x.reshape((batch_size, timesteps, num_input))
        # Run optimization op (backprop)
        sess.run(train_op, feed_dict={X: batch_x, Y: batch_y})
        if step % display_step == 0 or step == 1:
            # Calculate batch loss and accuracy
            loss, acc = sess.run([loss_op, accuracy], feed_dict={X: batch_x,
                                                                 Y: batch_y})
            print("Step " + str(step) + ", Minibatch Loss= " + 
                  "{:.4f}".format(loss) + ", Training Accuracy= " + 
                  "{:.3f}".format(acc))

    print("Optimization Finished!")

    # Calculate accuracy for 128 mnist test images
    test_len = 128
    test_data = mnist.test.images[:test_len].reshape((-1, timesteps, num_input))
    test_label = mnist.test.labels[:test_len]
    print("Testing Accuracy:", 
        sess.run(accuracy, feed_dict={X: test_data, Y: test_label}))

Step 1, Minibatch Loss= 2.8645, Training Accuracy= 0.062
Step 200, Minibatch Loss= 2.1180, Training Accuracy= 0.227
Step 400, Minibatch Loss= 1.9726, Training Accuracy= 0.344
Step 600, Minibatch Loss= 1.7784, Training Accuracy= 0.445
Step 800, Minibatch Loss= 1.5500, Training Accuracy= 0.547
Step 1000, Minibatch Loss= 1.5882, Training Accuracy= 0.453
Step 1200, Minibatch Loss= 1.5326, Training Accuracy= 0.555
Step 1400, Minibatch Loss= 1.3682, Training Accuracy= 0.570
Step 1600, Minibatch Loss= 1.3374, Training Accuracy= 0.594
Step 1800, Minibatch Loss= 1.1551, Training Accuracy= 0.648
Step 2000, Minibatch Loss= 1.2116, Training Accuracy= 0.633
Step 2200, Minibatch Loss= 1.1292, Training Accuracy= 0.609
Step 2400, Minibatch Loss= 1.0862, Training Accuracy= 0.680
Step 2600, Minibatch Loss= 1.0501, Training Accuracy= 0.672
Step 2800, Minibatch Loss= 1.0487, Training Accuracy= 0.688
Step 3000, Minibatch Loss= 1.0223, Training Accuracy= 0.727
Step 3200, Minibatch Loss= 1.0418, Training Accuracy= 0.695
Step 3400, Minibatch Loss= 0.8273, Training Accuracy= 0.719
Step 3600, Minibatch Loss= 0.9088, Training Accuracy= 0.727
Step 3800, Minibatch Loss= 0.9243, Training Accuracy= 0.750
Step 4000, Minibatch Loss= 0.8085, Training Accuracy= 0.703
Step 4200, Minibatch Loss= 0.8466, Training Accuracy= 0.711
Step 4400, Minibatch Loss= 0.8973, Training Accuracy= 0.734
Step 4600, Minibatch Loss= 0.7647, Training Accuracy= 0.750
Step 4800, Minibatch Loss= 0.9088, Training Accuracy= 0.742
Step 5000, Minibatch Loss= 0.7906, Training Accuracy= 0.742
Step 5200, Minibatch Loss= 0.7275, Training Accuracy= 0.781
Step 5400, Minibatch Loss= 0.7488, Training Accuracy= 0.789
Step 5600, Minibatch Loss= 0.7517, Training Accuracy= 0.758
Step 5800, Minibatch Loss= 0.7778, Training Accuracy= 0.797
Step 6000, Minibatch Loss= 0.6736, Training Accuracy= 0.742
Step 6200, Minibatch Loss= 0.6552, Training Accuracy= 0.773
Step 6400, Minibatch Loss= 0.5746, Training Accuracy= 0.828
Step 6600, Minibatch Loss= 0.8102, Training Accuracy= 0.727
Step 6800, Minibatch Loss= 0.6669, Training Accuracy= 0.773
Step 7000, Minibatch Loss= 0.6524, Training Accuracy= 0.766
Step 7200, Minibatch Loss= 0.6481, Training Accuracy= 0.805
Step 7400, Minibatch Loss= 0.5743, Training Accuracy= 0.828
Step 7600, Minibatch Loss= 0.6983, Training Accuracy= 0.773
Step 7800, Minibatch Loss= 0.5552, Training Accuracy= 0.828
Step 8000, Minibatch Loss= 0.5728, Training Accuracy= 0.820
Step 8200, Minibatch Loss= 0.5587, Training Accuracy= 0.789
Step 8400, Minibatch Loss= 0.5205, Training Accuracy= 0.836
Step 8600, Minibatch Loss= 0.4266, Training Accuracy= 0.906
Step 8800, Minibatch Loss= 0.7197, Training Accuracy= 0.812
Step 9000, Minibatch Loss= 0.4216, Training Accuracy= 0.852
Step 9200, Minibatch Loss= 0.4448, Training Accuracy= 0.844
Step 9400, Minibatch Loss= 0.3577, Training Accuracy= 0.891
Step 9600, Minibatch Loss= 0.4034, Training Accuracy= 0.883
Step 9800, Minibatch Loss= 0.4747, Training Accuracy= 0.828
Step 10000, Minibatch Loss= 0.5763, Training Accuracy= 0.805
Optimization Finished!
Testing Accuracy: 0.875