TensorFlow技术解析与实战 9 TensorFlow在MNIST中的应用

       MNIST (Mixed National Institute of Standards and Technology)是一个入门级的计算机视觉数据集,数据集中都是美国中学生手写的数字。它的训练集包含 6 万张图片,测试集包含 1万张图片,并且数字已经进行过预处理和格式化,做了大小调整并居中,图片尺寸也固定为28×28。这个数据集很小,但训练速度很快,而且收敛效果也很好,非常适合作为实战的例子

去学习。

9.1 MNIST 数据集简介

MNIST 数据集是 NIST 数据集的子集,包含以下 4 个文件。
    ● train-labels-idx1-ubyte.gz:训练集标记文件(28 881 字节)。
    ● train-images-idx3-ubyte.gz:训练集图片文件(9 912 422 字节)。
    ● t10k-labels-idx1-ubyte.gz:测试集标记文件(4 542 字节)。
    ● t10k-images-idx3-ubyte.gz:测试集图片文件(1 648 877 字节)。
MNIST 数据集包括训练集的图片和标记数据,以及测试集的图片和标记数据,在测试集包含的 10 000 个样例中,前 5 000 个样例取自原始的 NIST 训练集,后 5 000 个取自原始的 NIST测试集,因此前 5 000 个预测起来更容易些。

     下面具体讲解它们的格式http://yann.lecun.com/exdb/mnist

9.2 MNIST的分类问题

      Softmax回归可以解决两种以上的分类，该模型是Logistic回归模型的分类问题上的推广。

# -*- coding:utf-8 -*-import sysreload(sys)sys.setdefaultencoding('utf-8')from tensorflow.examples.tutorials.mnist import input_dataimport tensorflow as tf # 加载数据mnist = input_data.read_data_sets("/tmp/tensorflow/mnist/input_data/", one_hot=True)# 构建回归模型x = tf.placeholder(tf.float32, [None, 784])W = tf.Variable(tf.zeros([784, 10]))b = tf.Variable(tf.zeros([10]))y = tf.matmul(x, W) + b  # 预测值# 定义损失函数和优化器y_ = tf.placeholder(tf.float32, [None, 10])cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=y, labels=y_))# 采用SGD作为优化器train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)# 训练模型sess = tf.InteractiveSession()tf.global_variables_initializer().run()for _ in range(1000):batch_xs, batch_ys = mnist.train.next_batch(100)sess.run(train_step, feed_dict={x: batch_xs, y_:batch_ys})# 评估训练好的模型correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))  #计算预测值和真实值accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))  #布尔型转化为浮点数，并取平均值，得到准确率print(sess.run(accuracy, feed_dict={x:mnist.test.images, y_:mnist.test.labels}))   #计算模型在测试集上的准确率

0.918

9.3 训练过程的可视化

9.4 MNIST 的卷积神经网络

https://github.com/nlintz/TensorFlow-Tutorials/blob/master/05_convolutional_net.py

# -*- coding:utf-8 -*-import sysreload(sys)sys.setdefaultencoding('utf-8')from tensorflow.examples.tutorials.mnist import input_dataimport tensorflow as tf # 加载数据mnist = input_data.read_data_sets("", one_hot=True)trX, trY, teX, teY = mnist.train.images, mnist.train.labels, mnist.test.images, mnist.test.labelstrX = trX.reshape(-1, 28, 28, 1)   # 28x28x1 input imgteX = teX.reshape(-1, 28, 28, 1)   # 28x28x1 input imgX = tf.placeholder("float", [None, 28, 28, 1])Y = tf.placeholder("float", [None, 10])def init_weights(shape):return tf.Variable(tf.random_normal(shape, stddev=0.01))w = init_weights([3, 3, 1, 32])        # patch 大小为 3 × 3 ,输入维度为 1 ,输出维度为 32w2 = init_weights([3, 3, 32, 64])      # patch 大小为 3 × 3 ,输入维度为 32 ,输出维度为 64w3 = init_weights([3, 3, 64, 128])     # patch 大小为 3 × 3 ,输入维度为 64 ,输出维度为 128w4 = init_weights([128 * 4 * 4, 625])  # 全连接层,输入维度为 128 × 4 × 4, 是上一层的输出数据又三维的转变成一维, 输出维度为 625w_o = init_weights([625, 10])          # 输出层,输入维度为 625, 输出维度为 10 ,代表 10 类 (labels)def model(X, w, w2, w3, w4, w_o, p_keep_conv, p_keep_hidden):# 第一组卷积层及池化层l1a = tf.nn.relu(tf.nn.conv2d(X, w, strides=[1, 1, 1, 1], padding='SAME'))l1 = tf.nn.max_pool(l1a, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')l1 = tf.nn.dropout(l1, p_keep_conv)  # dropout 一些神经元# 第二组卷积层及池化层l2a = tf.nn.relu(tf.nn.conv2d(l1, w2, strides=[1, 1, 1, 1], padding='SAME'))l2 = tf.nn.max_pool(l2a, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')l2 = tf.nn.dropout(l2, p_keep_conv)  # dropout 一些神经元# 第三组卷积层及池化层l3a = tf.nn.relu(tf.nn.conv2d(l2, w3, strides=[1, 1, 1, 1], padding='SAME'))l3 = tf.nn.max_pool(l3a, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME') # l3 shape=(?, 4, 4, 128)l3 = tf.reshape(l3, [-1, w4.get_shape().as_list()[0]])  # reshape to (?, 128 * 4 * 4 = 2048)l3 = tf.nn.dropout(l3, p_keep_conv)# 全连接层，最后dropoutl4 = tf.nn.relu(tf.matmul(l3, w4))l4 = tf.nn.dropout(l4, p_keep_hidden)# 输出层pyx = tf.matmul(l4, w_o)return pyx   #返回预测值p_keep_conv = tf.placeholder("float")p_keep_hidden = tf.placeholder("float")py_x = model(X, w, w2, w3, w4, w_o, p_keep_conv, p_keep_hidden)# 定义损失函数cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=py_x, labels=Y))train_op = tf.train.RMSPropOptimizer(0.001, 0.9).minimize(cost)predict_op = tf.argmax(py_x, 1)#训练模型和评估模型batch_size = 128test_size = 256with tf.Session() as sess:tf.global_variables_initializer().run()for i in range(100):training_batch = zip(range(0, len(trX), batch_size), range(batch_size, len(trX)+1, batch_size))for start, end in training_batch:sess.run(train_op, feed_dict={X:trX[start:end], Y:trY[start:end], p_keep_conv:0.8, p_keep_hidden:0.5})test_indices = np.arange(len(teX)) np.random.shuffle(test_indices)test_indices = test_indices[0:test_size]print(i, np.mean(np.argmax(teY[test_indices], axis=1) ==              sess.run(predict_op, feed_dict={X: teX[test_indices],              p_keep_conv:1.0, p_keep_hidden: 1.0})))

9.5 MNIST 的循环神经网络

# -*- coding:utf-8 -*-import sysreload(sys)sys.setdefaultencoding('utf-8')from tensorflow.examples.tutorials.mnist import input_dataimport tensorflow as tf # 加载数据mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)trX, trY, teX, teY = mnist.train.images, mnist.train.labels, mnist.test.images, mnist.test.labelstrX = trX.reshape(-1, 28, 28, 1)   # 28x28x1 input imgteX = teX.reshape(-1, 28, 28, 1)   # 28x28x1 input img# 设置训练的超参数lr = 0.001training_iters = 100000batch_size = 128# 神经网络的参数n_inputs = 28     # 输入层的nn_steps = 28     # 28长度n_hidden_units = 128    # 隐藏层的神经元个数n_classes = 10  # 输出的数量，即分类的类别，0~9个数字，共有10个# 输入数据占位符x = tf.placeholder("float", [None, n_steps, n_inputs])y = tf.placeholder("float", [None, n_classes])# 定义权重weights = {'in': tf.Variable(tf.random_normal([n_inputs, n_hidden_units])),'out': tf.Variable(tf.random_normal([n_hidden_units, n_classes]))}biases = {'in': tf.Variable(tf.constant(0.1, shape=[n_hidden_units, ])),'out': tf.Variable(tf.constant(0.1, shape=[n_classes, ]))}#定义RNN模型def RNN(X, weights, biases):X = tf.reshape(X, [-1, n_inputs])  #把输入的X转换成X ==》（128 batch * 28 steps, 28 inputs）# 进入隐藏层X_in = tf.matmul(X, weights['in']) + biases['in']  # (128 batch * 28 steps, 128 hidden)X_in = tf.reshape(X_in, [-1, n_steps, n_hidden_units])  # 128 batch , 28 steps, 128 hidden# 这里采用基本的LSTM循环网络单元：basic LSTM Celllstm_cell = tf.contrib.rnn.BasicLSTMCell(n_hidden_units, forget_bias=1.0, state_is_tuple=True)init_state = lstm_cell.zero_state(batch_size, dtype=tf.float32) #lstm单元由两个部分组成：（c_state, h_state）# dynamic_rnn接收张量(batch, steps, inputs)或者（steps, batch, inputs）作为X_inoutputs, final_state = tf.nn.dynamic_rnn(lstm_cell, X_in, initial_state=init_state, time_major=False)results = tf.matmul(final_state[1], weights['out']) + biases['out']return results# 定义损失函数和优化器，优化器采用AdamOptimizerpred = RNN(x, weights, biases)cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pred, labels=y))train_op = tf.train.AdamOptimizer(lr).minimize(cost)# 定义模型预测结果及准确率计算方法correct_pred = tf.equal(tf.argmax(pred, 1), tf.argmax(y, 1))accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))# 训练数据及评估模型with tf.Session() as sess:sess.run(tf.global_variables_initializer())step = 0while step * batch_size < training_iters:batch_xs, batch_ys = mnist.train.next_batch(batch_size)batch_xs = batch_xs.reshape([batch_size, n_steps, n_inputs])sess.run([train_op], feed_dict={x:batch_xs, y:batch_ys,})if step % 20 == 0:print(sess.run(accuracy, feed_dict={x:batch_xs, y:batch_ys,}))step += 1

9.6 MNIST 的无监督学习

# -*- coding:utf-8 -*-import sysreload(sys)sys.setdefaultencoding('utf-8')import numpy as np import matplotlib.pyplot as pltfrom tensorflow.examples.tutorials.mnist import input_dataimport tensorflow as tf # 加载数据mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)trX, trY, teX, teY = mnist.train.images, mnist.train.labels, mnist.test.images, mnist.test.labelstrX = trX.reshape(-1, 28, 28, 1)   # 28x28x1 input imgteX = teX.reshape(-1, 28, 28, 1)   # 28x28x1 input imglearning_rate = 0.01   # 学习率training_epochs = 20   # 训练的轮数batch_size = 256     display_step = 1examples_to_show = 10n_hidden_1 = 256n_hidden_2 = 128n_input = 784X = tf.placeholder("float", [None, n_input])weights = {'encoder_h1': tf.Variable(tf.random_normal([n_input, n_hidden_1])),'encoder_h2': tf.Variable(tf.random_normal([n_hidden_1, n_hidden_2])),'decoder_h1': tf.Variable(tf.random_normal([n_hidden_2, n_hidden_1])),'decoder_h2': tf.Variable(tf.random_normal([n_hidden_1, n_input])),}biases = {'encoder_b1': tf.Variable(tf.random_normal([n_hidden_1])),'encoder_b2': tf.Variable(tf.random_normal([n_hidden_2])),'decoder_b1': tf.Variable(tf.random_normal([n_hidden_1])),'decoder_b2': tf.Variable(tf.random_normal([n_input])),}# 定义压缩函数def encoder(x):layer_1 = tf.nn.sigmoid(tf.add(tf.matmul(x, weights['encoder_h1']), biases['encoder_b1']))layer_2 = tf.nn.sigmoid(tf.add(tf.matmul(layer_1, weights['encoder_h2']), biases['encoder_b2']))return layer_2# 定义解压函数def decoder(x):layer_1 = tf.nn.sigmoid(tf.add(tf.matmul(x, weights['decoder_h1']), biases['decoder_b1']))layer_2 = tf.nn.sigmoid(tf.add(tf.matmul(layer_1, weights['decoder_h2']), biases['decoder_b2']))return layer_2# 构建模型encoder_op = encoder(X)decoder_op = decoder(encoder_op)# 得出预测值y_pred = decoder_op# 得出真实值，即输入值y_true = X# 定义损失函数和优化器cost = tf.reduce_mean(tf.pow(y_true - y_pred, 2))optimizer = tf.train.RMSPropOptimizer(learning_rate).minimize(cost)init = tf.global_variables_initializer()# 训练数据及评估模型with tf.Session() as sess:sess.run(init)total_batch = int(mnist.train.num_examples/batch_size)# 开始训练for epoch in range(training_epochs):for i in range(total_batch):batch_xs, batch_ys = mnist.train.next_batch(batch_size)_, c = sess.run([optimizer, cost], feed_dict={X:batch_xs})if epoch % display_step == 0:print("Epoch:", '%04d' % (epoch+1), "cost=", "{:.9f}".format(c))print("Optimization Finished!")# 对测试集应用训练好的自动编码网络encode_decode = sess.run(y_pred, feed_dict={X: mnist.test.images[:examples_to_show]})# 比较此时集原始图片和自动编码网络的重建结果f, a = plt.subplots(2, 10, figsize=(10, 2))for i in range(examples_to_show):a[0][i].imshow(np.reshape(mnist.test.images[i], (28, 28)))a[1][i].imshow(np.reshape(encode_decode[i], (28, 28)))  # 重建结果f.show()plt.draw()plt.waitforbuttonpress()