TensorFlow官方文档中文版-笔记(三)

增加隐含层实现MNIST任务

定义模型参数

in_units = 784# in_units是输入节点数h1_units = 300# h1_units是隐含层的输出节点数#初始化为截断的正态分布，其标准差为0.1 (因为模型使用的激活函数是Relu，所以需要使用正态分布给参数加点噪声，来打破完全对称并且避免0梯度)W1 = tf.Variable(tf.truncated_normal([in_units, h1_units], stddev=0.1))# 隐含层的权重b1 = tf.Variable(tf.zeros([h1_units]))# 隐含层的偏置W2 = tf.Variable(tf.zeros([h1_units, 10]))# 输出层的权重b2 = tf.Variable(tf.zeros([10]))# 输出层的偏置# 把Dropout的比率作为计算图的输入，并定义成一个placeholder，Dropout的比率keep_prob（保留节点的概率）是不一样的，通常在训练时小于1，预测时等于1，所以将比率作为计算图的输入keep_prob = tf.placeholder(tf.float32)

定义模型

# 定义模型结构# 定义隐含层，实现一个激活函数为ReLU的隐含层，y=relu(w1x+b1)hidden1 = tf.nn.relu(tf.matmul(x, W1) + b1)# 实现Dropout功能，即随机将一部分节点置为0，keep_prob参数为保留数据而不置为0的比例，训练时小于1，用于制造随机性防止过拟合，预测时等于1，使用全部特征来预测样本的类别hidden1_drop = tf.nn.dropout(hidden1, keep_prob)# 定义输出层y = tf.nn.softmax(tf.matmul(hidden1_drop, W2) + b2)

准确率为98%！

完整代码如下：

# TensorFlow进阶3—增加隐含层的神经网络import tensorflow as tffrom tensorflow.examples.tutorials.mnist import input_data# Import datamnist = input_data.read_data_sets("MNIST_data/", one_hot=True)sess = tf.InteractiveSession()# Step1、2 定义输入和定义算法公式in_units = 784# in_units是输入节点数h1_units = 300# h1_units是隐含层的输出节点数W1 = tf.Variable(tf.truncated_normal([in_units, h1_units], stddev=0.1))# 隐含层的权重b1 = tf.Variable(tf.zeros([h1_units]))# 隐含层的偏置W2 = tf.Variable(tf.zeros([h1_units, 10]))# 输出层的权重b2 = tf.Variable(tf.zeros([10]))# 输出层的偏置keep_prob = tf.placeholder(tf.float32)x = tf.placeholder(tf.float32, [None, in_units])hidden1 = tf.nn.relu(tf.matmul(x, W1) + b1)hidden1_drop = tf.nn.dropout(hidden1, keep_prob)y = tf.nn.softmax(tf.matmul(hidden1_drop, W2) + b2)# Step3 定义损失函数cross-entropyy_ = tf.placeholder(tf.float32, [None, 10])cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y), reduction_indices=[1]))# Step4 定义优化算法-SGD# 选择优化器（此处选择自适应优化器Adagrad）来优化losstrain_step = tf.train.AdagradOptimizer(0.3).minimize(cross_entropy)# Step5 迭代执行训练操作tf.global_variables_initializer().run()for i in range(3000):    batch_xs, batch_ys = mnist.train.next_batch(100)    train_step.run({x: batch_xs, y_: batch_ys, keep_prob: 0.75})# Step6 验证模型准确率correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))print(accuracy.eval({x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))