简单CNN实现MNIST手写数字集识别

取自于TensorFlow实战

from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
#设置TensorFlowCPU运算优先级
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
#导入手写数字输入集
mnist=input_data.read_data_sets('MNIST_data/',one_hot=True)
#定义交互式会话框
sess=tf.InteractiveSession()
#定义权重函数，截断正态分布
def weight_variable(shape):
    return tf.Variable(tf.truncated_normal(shape,stddev=0.1))
#定义偏置函数
def bias_variable(shape):
    return tf.Variable(tf.constant(0.1,shape=shape))
#定义二维卷积
def conv2d(x,w):
    return tf.nn.conv2d(x,w,strides=[1,1,1,1],padding='SAME')
#定义2x2池化
def max_pool_2x2(x):
    return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')

#定义输入，标签、dropout保留率占位符
x=tf.placeholder(tf.float32,[None,784])
y_=tf.placeholder(tf.float32,[None,10])
keep_prob=tf.placeholder(tf.float32)
#就输入转化为图像形式
x_image=tf.reshape(x,[-1,28,28,1])
#第一卷积层，卷积、激活、池化
w_conv1=weight_variable([5,5,1,32])
b_conv1=bias_variable([32])
h_conv1=tf.nn.relu(conv2d(x_image,w_conv1)+b_conv1)
h_pool1=max_pool_2x2(h_conv1)
#第二卷积层，卷积、激活、池化
w_conv2=weight_variable([5,5,32,64])
b_conv2=bias_variable([64])
h_conv2=tf.nn.relu(conv2d(h_pool1,w_conv2)+b_conv2)
h_pool2=max_pool_2x2(h_conv2)
#第一全连接层，一维化、线性、激活、dropout
w_fc1=weight_variable([7*7*64,1024])
b_fc1=bias_variable([1024])
h_pool2_flat=tf.reshape(h_pool2,[-1,7*7*64])
h_fc1=tf.nn.relu(tf.matmul(h_pool2_flat,w_fc1)+b_fc1)
h_fc1_drop=tf.nn.dropout(h_fc1,keep_prob)
#第二全连接层，线性、激活
w_fc2=weight_variable([1024,10])
b_fc2=bias_variable([10])
y=tf.nn.softmax(tf.matmul(h_fc1_drop,w_fc2)+b_fc2)
#定义损失函数为交叉熵
cross_entropy=tf.reduce_mean(-tf.reduce_sum(y_*tf.log(y),reduction_indices=[1]))
#使用亚当优化器，最小化交叉熵
train=tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
#定义预测正误判断模型，（模型最大值index，标签最大值index）是否相等
correct_prediction=tf.equal(tf.argmax(y,1),tf.argmax(y_,1))
#计算预测准确率，将预测正误判断模型的bool型转化为float32,对其求平均为准确率
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
#初始化计算图
tf.global_variables_initializer().run()
#定义训练循环，minibatch size=50，训练keep_prob=0.75,预测keep_prob=1
for i in range(20000):
    batch=mnist.train.next_batch(50)
    train.run({x:batch[0],y_:batch[1],keep_prob:0.75})
    if i%100==0:
        train_accuracy=accuracy.eval({x:batch[0],y_:batch[1],keep_prob:1})
        print('step{:5d}:rain_accuracy={:g}'.format(i,train_accuracy))
#使用eval获取test_accuracy
print('test_accuracy=%g'%accuracy.eval({x:mnist.test.images,y_:mnist.test.labels,keep_prob:1}))