keras框架的CNN手写数字识别MNIST

参考：林大贵.TensorFlow+Keras深度学习人工智能实践应用[M].北京：清华大学出版社,2018.

首先在命令行中写入 activate tensorflow和jupyter notebook，运行如下代码。当然，事先准备好MNIST数据集。

 # coding: utf-8

 # In[4]:

 from keras.datasets import mnist
 from keras.utils import np_utils
 import numpy as np
 np.random.seed(10)

 # In[5]:

 (x_train,y_train),(x_test,y_test)=mnist.load_data()

 # In[6]:

 x_train4d = x_train.reshape(x_train.shape[0],28,28,1).astype('float32')
 x_test4d = x_test.reshape(x_test.shape[0],28,28,1).astype('float32')

 # In[7]:

 x_train4d_normalize = x_train4d/255
 x_test4d_normalize = x_test4d/255

 # In[8]:

 y_train_oneHot = np_utils.to_categorical(y_train)
 y_test_oneHot = np_utils.to_categorical(y_test)

 # In[9]:

 from keras.models import Sequential
 from keras.layers import Dense,Dropout,Flatten,Conv2D,MaxPooling2D

 # In[10]:

 model = Sequential()

 # In[11]:

 model.add(Conv2D(filters = 16,
                  kernel_size = (5,5),
                  padding = 'same',
                  input_shape = (28,28,1),
                  activation = ('relu')
                 ))

 # In[12]:

 model.add(MaxPooling2D(pool_size=(2,2)))

 # In[13]:

 model.add(Conv2D(filters = 36,
                 kernel_size = (5,5),
                 padding = 'same',
                 activation = 'relu'))

 # In[14]:

 model.add(MaxPooling2D(pool_size=(2,2)))

 # In[15]:

 model.add(Dropout(0,255))

 # In[16]:

 model.add(Flatten())

 # In[17]:

 model.add(Dense(128,activation = 'relu'))

 # In[18]:

 model.add(Dropout(0.5))

 # In[19]:

 model.add(Dense(10,activation = 'sigmoid'))

 # In[20]:

 print(model.summary())

 # In[21]:

 model.compile(loss='categorical_crossentropy',
              optimizer = 'adam',
              metrics = ['accuracy'])

 # In[22]:

 train_history = model.fit(x = x_train4d_normalize,
                          y = y_train_oneHot,
                          validation_split = 0.2,
                          epochs = 10,
                          batch_size = 300,
                          verbose = 2)

 # In[23]:

 import matplotlib.pyplot as plt
 def show_train_history(train_history,train,validation):
     plt.plot(train_history.history[train])
     plt.plot(train_history.history[validation])
     plt.title('Train_History')
     plt.ylabel(train)
     plt.xlabel('Epoch')
     plt.legend(['train','validation'], loc = 'upper left')
     plt.show()

 # In[24]:

 show_train_history(train_history,'acc','val_acc')

 # In[25]:

 show_train_history(train_history,'loss','val_loss')

 # In[26]:

 scores = model.evaluate(x_test4d_normalize,y_test_oneHot)
 scores[1]

 # In[27]:

 def plot_image_labels_prediction(images,labels,prediction,idx,num=10):
     fig = plt.gcf()
     fig.set_size_inches(12,24)
     if num>50 : num = 50
     for i in range(0,num):
         ax = plt.subplot(10,5,1+i)
         ax.imshow(images[idx],cmap='binary')
         title = "lable="+str(labels[idx])
         if len(prediction)>0:
             title+=",predict="+str(prediction[idx])
         ax.set_title(title,fontsize=10)
         ax.set_xticks([]);ax.set_yticks([])
         idx+=1
     plt.show()                                  

 # In[28]:

 prediction = model.predict_classes(x_test4d_normalize)

 # In[29]:

 plot_image_labels_prediction(x_test,
                              y_test,
                              prediction,
                              0,
                             50)

 # In[30]:

 import pandas as pd

 # In[31]:

 pd.crosstab(y_test,
             prediction,
            rownames=['label'],
            colnames=['predict'])

 # In[ ]:

卷积神经网络简介：https://www.cnblogs.com/bai2018/p/10413889.html

产生如下神经网络：

其中加入的dropout层用于避免过度拟合。在每次迭代时，随机舍弃一部分的训练样本。

训练效果较好。

林大贵.TensorFlow+Keras深度学习人工智能实践应用[M].北京：清华大学出版社,2018.