%matplotlib inline
import tensorflow as tf
import tensorflow.examples.tutorials.mnist.input_data as input_data
mnist=input_data.read_data_sets('MNIST_data/',one_hot=True)
import matplotlib.pyplot as plt
 
def plot_image(image):                #图片显示函数
  plt.imshow(image.reshape(28,28),cmap='binary')
  plt.show()
 
print("训练集数量：",mnist.train.num_examples,
   "特征值组成：",mnist.train.images.shape,
   "标签组成：",mnist.train.labels.shape)
 
batch_images,batch_labels=mnist.train.next_batch(batch_size=10)  #批量读取数据
print(batch_images.shape,batch_labels.shape)
 
print('标签值：',np.argmax(mnist.train.labels[1000]),end=' ')  #np.argmax()得到实际值
print('独热编码表示：',mnist.train.labels[1000])
plot_image(mnist.train.images[1000])         #显示数据集中第1000张图片

import numpy as np
import tensorflow as tf
import tensorflow.examples.tutorials.mnist.input_data as input_data
mnist=input_data.read_data_sets('MNIST_data/',one_hot=True)
import matplotlib.pyplot as plt
 
#定义占位符、变量、前向计算
x=tf.placeholder(tf.float32,[None,784],name='x')
y=tf.placeholder(tf.float32,[None,10],name='y')
W=tf.Variable(tf.random_normal([784,10]),name='W')
b=tf.Variable(tf.zeros([10]),name='b')
forward=tf.matmul(x,W)+b
pred=tf.nn.softmax(forward)               #通过softmax将线性结果分类处理
 
#计算预测值与真实值的匹配个数
correct_prediction=tf.equal(tf.argmax(pred,1),tf.argmax(y,1))
#将上一步得到的布尔值转换为浮点数，并求平均值，得到准确率
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
 
ss=tf.Session()
init=tf.global_variables_initializer()
ss.run(init)
 
#超参数设置
train_epochs=50
batch_size=100                        #每个批次的样本数
batch_num=int(mnist.train.num_examples/batch_size)      #一轮需要训练多少批
learning_rate=0.01
 
#定义交叉熵损失函数、梯度下降优化器
loss_function=tf.reduce_mean(-tf.reduce_sum(y*tf.log(pred),reduction_indices=1)) 
optimizer=tf.train.GradientDescentOptimizer(learning_rate).minimize(loss_function)
 
for epoch in range(train_epochs):
  for batch in range(batch_num):              #分批次读取数据进行训练
    xs,ys=mnist.train.next_batch(batch_size)
    ss.run(optimizer,feed_dict={x:xs,y:ys})
  #每轮训练结束后通过带入验证集的数据，检测模型的损失与准去率 
  loss,acc=ss.run([loss_function,accuracy],          feed_dict={x:mnist.validation.images,y:mnist.validation.labels})
  print('第%2d轮训练：损失为：%9f，准确率：%.4f'%(epoch+1,loss,acc))

prediction=ss.run(tf.argmax(pred,1),feed_dict={x:mnist.test.images})
 
def show_result(images,labels,prediction,index,num=10):   #绘制图形显示预测结果
  pic=plt.gcf()                      #获取当前图像
  pic.set_size_inches(10,12)               #设置图片大小
  for i in range(0,num):
    sub_pic=plt.subplot(5,5,i+1)            #获取第i个子图
    #将第index个images信息显示到子图上
    sub_pic.imshow(np.reshape(images[index],(28,28)),cmap='binary') 
    title="label:"+str(np.argmax(labels[index]))    #设置子图的title内容
    if len(prediction)>0:
      title+=",predict:"+str(prediction[index])
      
    sub_pic.set_title(title,fontsize=10)
    sub_pic.set_xticks([])               #设置x、y坐标轴不显示
    sub_pic.set_yticks([])
    index+=1
  plt.show()
show_result(mnist.test.images,mnist.test.labels,prediction,10)