原文連結：https://data-flair.training/blogs/python-deep-learning-project-handwritten-digit-recognition/

原文講得很詳細，這裡補充一些註釋。由於直接從庫匯入mnist資料集需要的時間非常久，因此這裡匯入的是本地已下載好的mnist資料集。（但我懷疑我下了假的資料集，咋驗證準確率這麼低，所以這裡不提供了）

import keras
from keras import backend as K
import numpy as np
from keras.datasets import mnist
from keras.models import
 
 Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D

batch_size = 128   #一次訓練所選取的樣本數
num_classes = 10   #分類個數
epochs = 10        #訓練輪數


#讀取已下載到本地的資料集
f=np.load('C:/Users/Administrator/.keras/datasets/mnist.npz')
x_train,y_train=f['x_train'],f['y_train
 
']
x_test,y_test=f['x_test'],f['y_test']
#print(x_train.shape, y_train.shape)


#資料預處理
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
x_test = x_test.reshape(x_test.shape[0], 28, 28, 1)
input_shape = (28, 28, 1)

x_train = x_train.astype('float32')  #轉換資料型別
x_test = x_test.astype('float32')
x_train 
 
/= 255      #歸一化
x_test /= 255 
y_train = keras.utils.to_categorical(y_train, num_classes)   #將整形陣列轉化為二元型別矩陣
y_test = keras.utils.to_categorical(y_test, num_classes)
#print('x_train shape:', x_train.shape)
#print(x_train.shape[0], 'train samples')
#print(x_test.shape[0], 'test samples')


#建立CNN模型
model = Sequential()  #這裡採用順序模型構建CNN
#輸入層，這裡指定輸入資料形狀為28*28*1 卷積核數量為32 形狀為3*3
model.add(Conv2D(32, kernel_size=(3, 3),activation='relu',input_shape=input_shape))
#新增中間層
model.add(Conv2D(64, (3, 3), activation='relu'))   #卷積層
model.add(MaxPooling2D(pool_size=(2, 2)))          #最大池化層
model.add(Dropout(0.25))                           #通過Dropout防止過擬合
model.add(Flatten())                               #展平層
model.add(Dense(256, activation='relu'))           #全連線層
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))
#損失函式
model.compile(loss=keras.losses.categorical_crossentropy,optimizer=keras.optimizers.Adadelta(),metrics=['accuracy'])

#訓練模型
hist = model.fit(x_train, y_train,batch_size=batch_size,epochs=epochs,verbose=2,validation_data=(x_test, y_test))
print("模型訓練完成")

#模型評估
score = model.evaluate(x_test, y_test, verbose=0)
print('test loss: ', score[0])
print('test accuracy: ', score[1])