下载MNIST数据集（28*28，输入维度为784）
import tensorflow as tf#下载MNIST数据集（28*28，输入维度为784）from tensorflow.examples.tutorials.mnist import input_datamnist = input_data.read_data_sets('MNIST_data', one_hot=True)一、定义Weight变量、biase变量、卷积层、池化层
#定义Weight变量，输入shape，返回变量的参数#随机初始化,tf.truncated_normal(shape, mean=0.0, stddev(标准差)=1.0, dtype=tf.float32, seed=None, name=None)def weight_variable(shape): init = tf.truncated_normal(shape, stddev=0.1) return tf.Variable(init)#定义biase变量，传入shape ,返回变量的一些参数#用不为0的数初始化def weight_biase(shape): init = tf.constant(0.1,shape = shape) return tf.Variable(init)#定义卷积，传入x和参数，Wtf.nn.conv2d(input, filter, strides, padding, use_cudnn_on_gpu=None, name=None)def conv2d(x,W): return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')#stride的第一位和第四位默认为1，第二、三位为x,y移动的距离#定义池化，传入x,池化层无参数，pooling,tf.nn.max_pool(value, ksize, strides, padding, name=None)def max_poo_2x2(x): return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding = 'SAME')#ksize为卷积核大小2*2，第一位和第四位默认为2#stride的第一位和第四位默认为1，第二、三位为x,y移动的距离#定义一个计算准确率的函数def compute_accuracy(v_xs, v_ys): global prediction y_pre = sess.run(prediction, feed_dict={xs: v_xs,keep_prob: 1}) correct_prediction = tf.equal(tf.argmax(y_pre,1), tf.argmax(v_ys,1)) accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) result = sess.run(accuracy, feed_dict={xs: v_xs, ys: v_ys,keep_prob: 1}) return result
#定义输入的placeholderxs = tf.placeholder(tf.float32, [None, 784])#图片28*28维ys = tf.placeholder(tf.float32, [None, 10])keep_prob = tf.placeholder(tf.float32)#将xs的形状reshape为[-1,28,28,1]，-1为任意sample数，1为chanel数，图片为28,28,1x_image = tf.reshape(xs,[-1,28,28,1])#第一层：convolutional layer1 + max pooling+reluW_conv1 = weight_variable([5,5,1,32])#filter大小为5*5，channel=1，32个filterb_conv1 = weight_biase([32])#注意：输入conv1的卷积层是x_image，是reshape后的28,28,1h_conv1 = tf.nn.relu(conv2d(x_image,W_conv1)+b_conv1)h_pool1 = max_poo_2x2(h_conv1)#第二层：convolutional layer2 + max pooling+reluW_conv2 = weight_variable([5,5,32,64])#filter大小为5*5，channel=32，64个filterb_conv2 = weight_biase([64])#注意：输入conv1的卷积层是x_image，是reshape后的28,28,1h_conv2 = tf.nn.relu(conv2d(h_pool1,W_conv2)+b_conv2)h_pool2 = max_poo_2x2(h_conv2)#第三层：fully connected layer1 +relu+ dropout#注意：将输出的7,7,64展开为一维向量！！！h_pool2 = tf.reshape(h_pool2,[-1,7*7*64])W_fc1 = weight_variable([7*7*64,1024])b_fc1 = weight_biase([1024])h_fc1 = tf.nn.relu(tf.matmul(h_pool2,W_fc1)+b_fc1)#考虑过拟合问题，加入一个dropout处理h_fc1_drop = tf.nn.dropout(h_fc1,keep_prob)#第四层：fully connected layer2 to prediction+softmax分类W_fc2 = weight_variable([1024,10])b_fc2 = weight_biase([10])prediction = tf.nn.softmax(tf.matmul(h_fc1_drop,W_fc2)+b_fc2)#损失函数(cross_entropy)和优化方法cross_entropy = tf.reduce_mean(-tf.reduce_sum(ys*tf.log(prediction), reduction_indices=[1]))train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)#session会话控制sess = tf.Session()sess.run(tf.global_variables_initializer())#训练1000次，每训练50次输出测试数据的训练精度for i in range(1000): #开始训练，训练集中每次取100个数据（batch_xs, batch_ys） batch_xs, batch_ys = mnist.train.next_batch(100) sess.run(train_step, feed_dict={xs: batch_xs, ys: batch_ys, keep_prob: 0.5})#placeholder和feed_dict同时出现 if i%50 == 0: print(compute_accuracy( mnist.test.images, mnist.test.labels))
结果：（跑的时间太长，提前暂停了，可以看到第二个epoch精确度就达到了0.79）