静态多层LSTM对MNIST分类

# -*- coding: utf-8 -*-
import numpy as np
import tensorflow as tf
# 导入 MINST 数据集
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/data/", one_hot=True)

n_input = 28 # MNIST data 输入 (img shape: 28*28)
n_steps = 28 # timesteps
n_hidden = 128 # hidden layer num of features
n_classes = 10  # MNIST 列别 (0-9 ，一共10类)
batch_size = 128


tf.reset_default_graph()
# tf Graph input
x = tf.placeholder("float", [None, n_steps, n_input])
y = tf.placeholder("float", [None, n_classes])



stacked_rnn = []
for i in range(3):
    stacked_rnn.append(tf.contrib.rnn.LSTMCell(n_hidden))
mcell = tf.contrib.rnn.MultiRNNCell(stacked_rnn)

x1 = tf.unstack(x, n_steps, 1)
outputs, states = tf.contrib.rnn.static_rnn(mcell, x1, dtype=tf.float32)
pred = tf.contrib.layers.fully_connected(outputs[-1],n_classes,activation_fn = None)


learning_rate = 0.001
# Define loss and optimizer
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y))
optimizer = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(cost)

# Evaluate model
correct_pred = tf.equal(tf.argmax(pred,1), tf.argmax(y,1))
accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))


training_iters = 100000

display_step = 10

# 启动session
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    step = 1
    # Keep training until reach max iterations
    while step * batch_size < training_iters:
        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, n_steps, n_input))
        # Run optimization op (backprop)
        sess.run(optimizer, feed_dict={x: batch_x, y: batch_y})
        if step % display_step == 0:
            # 计算批次数据的准确率
            acc = sess.run(accuracy, feed_dict={x: batch_x, y: batch_y})
            # Calculate batch loss
            loss = sess.run(cost, feed_dict={x: batch_x, y: batch_y})
            print ("Iter " + str(step*batch_size) + ", Minibatch Loss= " + \
                  "{:.6f}".format(loss) + ", Training Accuracy= " + \
                  "{:.5f}".format(acc))
        step += 1
    print (" Finished!")

    # 计算准确率 for 128 mnist test images
    test_len = 100
    test_data = mnist.test.images[:test_len].reshape((-1, n_steps, n_input))
    test_label = mnist.test.labels[:test_len]
    print ("Testing Accuracy:", \
        sess.run(accuracy, feed_dict={x: test_data, y: test_label}))

实验结果：
Iter 1280, Minibatch Loss= 1.949785, Training Accuracy= 0.28125
Iter 2560, Minibatch Loss= 1.590585, Training Accuracy= 0.44531
Iter 3840, Minibatch Loss= 1.219217, Training Accuracy= 0.57031
Iter 5120, Minibatch Loss= 0.783433, Training Accuracy= 0.74219
Iter 6400, Minibatch Loss= 0.680670, Training Accuracy= 0.81250
Iter 7680, Minibatch Loss= 0.732436, Training Accuracy= 0.79688
Iter 8960, Minibatch Loss= 0.647141, Training Accuracy= 0.82031
Iter 10240, Minibatch Loss= 0.440039, Training Accuracy= 0.84375
Iter 11520, Minibatch Loss= 0.411871, Training Accuracy= 0.90625
Iter 12800, Minibatch Loss= 0.717502, Training Accuracy= 0.78125
Iter 14080, Minibatch Loss= 0.426217, Training Accuracy= 0.89062
Iter 15360, Minibatch Loss= 0.253466, Training Accuracy= 0.92188
Iter 16640, Minibatch Loss= 0.294188, Training Accuracy= 0.90625
Iter 17920, Minibatch Loss= 0.438532, Training Accuracy= 0.86719
Iter 19200, Minibatch Loss= 0.264974, Training Accuracy= 0.93750
Iter 20480, Minibatch Loss= 0.257035, Training Accuracy= 0.91406
Iter 21760, Minibatch Loss= 0.291660, Training Accuracy= 0.89062
Iter 23040, Minibatch Loss= 0.322604, Training Accuracy= 0.90625
Iter 24320, Minibatch Loss= 0.284016, Training Accuracy= 0.91406
Iter 25600, Minibatch Loss= 0.275679, Training Accuracy= 0.91406
Iter 26880, Minibatch Loss= 0.287420, Training Accuracy= 0.92969
Iter 28160, Minibatch Loss= 0.239718, Training Accuracy= 0.93750
Iter 29440, Minibatch Loss= 0.241036, Training Accuracy= 0.94531
Iter 30720, Minibatch Loss= 0.384207, Training Accuracy= 0.91406
Iter 32000, Minibatch Loss= 0.308553, Training Accuracy= 0.89844
Iter 33280, Minibatch Loss= 0.326937, Training Accuracy= 0.88281
Iter 34560, Minibatch Loss= 0.201968, Training Accuracy= 0.92969
Iter 35840, Minibatch Loss= 0.168276, Training Accuracy= 0.93750
Iter 37120, Minibatch Loss= 0.255253, Training Accuracy= 0.91406
Iter 38400, Minibatch Loss= 0.143009, Training Accuracy= 0.96875
Iter 39680, Minibatch Loss= 0.131618, Training Accuracy= 0.96875
Iter 40960, Minibatch Loss= 0.109194, Training Accuracy= 0.96094
Iter 42240, Minibatch Loss= 0.243728, Training Accuracy= 0.92188
Iter 43520, Minibatch Loss= 0.129906, Training Accuracy= 0.97656
Iter 44800, Minibatch Loss= 0.100776, Training Accuracy= 0.96875
Iter 46080, Minibatch Loss= 0.210856, Training Accuracy= 0.93750
Iter 47360, Minibatch Loss= 0.172149, Training Accuracy= 0.96875
Iter 48640, Minibatch Loss= 0.201521, Training Accuracy= 0.93750
Iter 49920, Minibatch Loss= 0.150544, Training Accuracy= 0.93750
Iter 51200, Minibatch Loss= 0.146078, Training Accuracy= 0.92188
Iter 52480, Minibatch Loss= 0.107032, Training Accuracy= 0.96094
Iter 53760, Minibatch Loss= 0.076864, Training Accuracy= 0.97656
Iter 55040, Minibatch Loss= 0.063498, Training Accuracy= 0.99219
Iter 56320, Minibatch Loss= 0.161256, Training Accuracy= 0.95312
Iter 57600, Minibatch Loss= 0.173243, Training Accuracy= 0.94531
Iter 58880, Minibatch Loss= 0.139156, Training Accuracy= 0.96094
Iter 60160, Minibatch Loss= 0.086291, Training Accuracy= 0.96875
Iter 61440, Minibatch Loss= 0.160935, Training Accuracy= 0.96094
Iter 62720, Minibatch Loss= 0.105631, Training Accuracy= 0.95312
Iter 64000, Minibatch Loss= 0.114493, Training Accuracy= 0.96094
Iter 65280, Minibatch Loss= 0.175501, Training Accuracy= 0.96094
Iter 66560, Minibatch Loss= 0.081360, Training Accuracy= 0.97656
Iter 67840, Minibatch Loss= 0.134259, Training Accuracy= 0.96094
Iter 69120, Minibatch Loss= 0.045177, Training Accuracy= 0.98438
Iter 70400, Minibatch Loss= 0.088118, Training Accuracy= 0.97656
Iter 71680, Minibatch Loss= 0.157305, Training Accuracy= 0.95312
Iter 72960, Minibatch Loss= 0.079270, Training Accuracy= 0.97656
Iter 74240, Minibatch Loss= 0.216380, Training Accuracy= 0.94531
Iter 75520, Minibatch Loss= 0.201112, Training Accuracy= 0.95312
Iter 76800, Minibatch Loss= 0.070850, Training Accuracy= 0.97656
Iter 78080, Minibatch Loss= 0.147870, Training Accuracy= 0.95312
Iter 79360, Minibatch Loss= 0.078924, Training Accuracy= 0.98438
Iter 80640, Minibatch Loss= 0.046514, Training Accuracy= 0.99219
Iter 81920, Minibatch Loss= 0.048644, Training Accuracy= 0.98438
Iter 83200, Minibatch Loss= 0.066273, Training Accuracy= 0.97656
Iter 84480, Minibatch Loss= 0.083965, Training Accuracy= 0.97656
Iter 85760, Minibatch Loss= 0.066841, Training Accuracy= 0.97656
Iter 87040, Minibatch Loss= 0.067535, Training Accuracy= 0.99219
Iter 88320, Minibatch Loss= 0.161641, Training Accuracy= 0.96875
Iter 89600, Minibatch Loss= 0.073918, Training Accuracy= 0.97656
Iter 90880, Minibatch Loss= 0.056344, Training Accuracy= 0.98438
Iter 92160, Minibatch Loss= 0.121526, Training Accuracy= 0.95312
Iter 93440, Minibatch Loss= 0.061281, Training Accuracy= 0.97656
Iter 94720, Minibatch Loss= 0.149234, Training Accuracy= 0.94531
Iter 96000, Minibatch Loss= 0.150431, Training Accuracy= 0.93750
Iter 97280, Minibatch Loss= 0.094178, Training Accuracy= 0.97656
Iter 98560, Minibatch Loss= 0.052903, Training Accuracy= 0.98438
Iter 99840, Minibatch Loss= 0.077210, Training Accuracy= 0.97656
Finished!
Testing Accuracy: 0.98

---

参考资料：《深度学习之Tensorflow》李金洪编著