from __future__ import print_function
# 导入numpy库， numpy是一个常用的科学计算库，优化矩阵的运算
import numpy as np
np.random.seed(1337)

# 导入mnist数据库， mnist是常用的手写数字库
from keras.datasets import mnist
# 导入顺序模型
from keras.models import Sequential
# 导入全连接层Dense， 激活层Activation 以及 Dropout层
from keras.layers.core import Dense, Dropout, Activation
# 导入优化器RMSProp
from keras.optimizers import RMSprop
# 导入numpy工具，主要是用to_categorical来转换类别向量
from keras.utils import np_utils

# 设置batch的大小
batch_size = 128
# 设置类别的个数
nb_classes = 10
# 设置迭代的次数
nb_epoch = 20

# keras中的mnist数据集已经被划分成了60,000个训练集，10,000个测试集的形式，按以下格式调用即可
(X_train, y_train), (X_test, y_test) = mnist.load_data()

# X_train原本是一个60000*28*28的三维向量，将其转换为60000*784的二维向量
X_train = X_train.reshape(60000, 784)
# X_test原本是一个10000*28*28的三维向量，将其转换为10000*784的二维向量
X_test = X_test.reshape(10000, 784)
# 将X_train, X_test的数据格式转为float32存储
X_train = X_train.astype('float32')
X_test = X_test.astype('float32')
# 归一化
X_train /= 255
X_test /= 255
# 打印出训练集和测试集的信息
print(X_train.shape[0], 'train samples')
print(X_test.shape[0], 'test samples')

60000 train samples
10000 test samples

'''
将类别向量(从0到nb_classes的整数向量)映射为二值类别矩阵，
相当于将向量用one-hot重新编码'''
Y_train = np_utils.to_categorical(y_train, nb_classes)
Y_test = np_utils.to_categorical(y_test, nb_classes)

# 建立顺序型模型
model = Sequential()
'''
模型需要知道输入数据的shape，
因此，Sequential的第一层需要接受一个关于输入数据shape的参数，
后面的各个层则可以自动推导出中间数据的shape，
因此不需要为每个层都指定这个参数
''' 

# 输入层有784个神经元
# 第一个隐层有512个神经元，激活函数为ReLu，Dropout比例为0.2
model.add(Dense(512, input_shape=(784,)))
model.add(Activation('relu'))
model.add(Dropout(0.2))

# 第二个隐层有512个神经元，激活函数为ReLu，Dropout比例为0.2
model.add(Dense(512))
model.add(Activation('relu'))
model.add(Dropout(0.2))

# 输出层有10个神经元，激活函数为SoftMax，得到分类结果
model.add(Dense(10))
model.add(Activation('softmax'))

# 输出模型的整体信息
# 总共参数数量为784*512+512 + 512*512+512 + 512*10+10 = 669706
model.summary()

____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
dense_4 (Dense)                  (None, 512)           401920      dense_input_2[0][0]              
____________________________________________________________________________________________________
activation_4 (Activation)        (None, 512)           0           dense_4[0][0]                    
____________________________________________________________________________________________________
dropout_3 (Dropout)              (None, 512)           0           activation_4[0][0]               
____________________________________________________________________________________________________
dense_5 (Dense)                  (None, 512)           262656      dropout_3[0][0]                  
____________________________________________________________________________________________________
activation_5 (Activation)        (None, 512)           0           dense_5[0][0]                    
____________________________________________________________________________________________________
dropout_4 (Dropout)              (None, 512)           0           activation_5[0][0]               
____________________________________________________________________________________________________
dense_6 (Dense)                  (None, 10)            5130        dropout_4[0][0]                  
____________________________________________________________________________________________________
activation_6 (Activation)        (None, 10)            0           dense_6[0][0]                    
====================================================================================================
Total params: 669,706
Trainable params: 669,706
Non-trainable params: 0
____________________________________________________________________________________________________

'''
配置模型的学习过程
compile接收三个参数：
1.优化器optimizer：参数可指定为已预定义的优化器名，如rmsprop、adagrad，
或一个Optimizer类对象，如此处的RMSprop()
2.损失函数loss：参数为模型试图最小化的目标函数，可为预定义的损失函数，
如categorical_crossentropy、mse，也可以为一个损失函数
3.指标列表：对于分类问题，一般将该列表设置为metrics=['accuracy']
'''
model.compile(loss='categorical_crossentropy',
              optimizer=RMSprop(),
              metrics=['accuracy'])

'''
训练模型
batch_size：指定梯度下降时每个batch包含的样本数
nb_epoch：训练的轮数，nb指number of
verbose：日志显示，0为不在标准输出流输出日志信息，1为输出进度条记录，2为epoch输出一行记录
validation_data：指定验证集
fit函数返回一个History的对象，其History.history属性记录了损失函数和其他指标的数值随epoch变化的情况，
如果有验证集的话，也包含了验证集的这些指标变化情况
'''
history = model.fit(X_train, Y_train,
                    batch_size = batch_size,
                    nb_epoch = nb_epoch,
                    verbose = 1,
                    validation_data = (X_test, Y_test))

# 按batch计算在某些输入数据上模型的误差
score = model.evaluate(X_test, Y_test, verbose=0)

Train on 60000 samples, validate on 10000 samples
Epoch 1/20
60000/60000 [==============================] - 3s - loss: 0.2468 - acc: 0.9245 - val_loss: 0.1062 - val_acc: 0.9662
Epoch 2/20
60000/60000 [==============================] - 3s - loss: 0.1027 - acc: 0.9687 - val_loss: 0.0885 - val_acc: 0.9744
Epoch 3/20
60000/60000 [==============================] - 3s - loss: 0.0755 - acc: 0.9772 - val_loss: 0.0798 - val_acc: 0.9763
Epoch 4/20
60000/60000 [==============================] - 3s - loss: 0.0617 - acc: 0.9810 - val_loss: 0.1023 - val_acc: 0.9692
Epoch 5/20
60000/60000 [==============================] - 3s - loss: 0.0512 - acc: 0.9847 - val_loss: 0.0832 - val_acc: 0.9791
Epoch 6/20
60000/60000 [==============================] - 3s - loss: 0.0447 - acc: 0.9866 - val_loss: 0.0778 - val_acc: 0.9796
Epoch 7/20
60000/60000 [==============================] - 3s - loss: 0.0392 - acc: 0.9883 - val_loss: 0.0822 - val_acc: 0.9798
Epoch 8/20
60000/60000 [==============================] - 3s - loss: 0.0336 - acc: 0.9899 - val_loss: 0.0784 - val_acc: 0.9820
Epoch 9/20
60000/60000 [==============================] - 3s - loss: 0.0336 - acc: 0.9904 - val_loss: 0.0937 - val_acc: 0.9809
Epoch 10/20
60000/60000 [==============================] - 3s - loss: 0.0293 - acc: 0.9917 - val_loss: 0.0802 - val_acc: 0.9829
Epoch 11/20
60000/60000 [==============================] - 3s - loss: 0.0260 - acc: 0.9924 - val_loss: 0.0966 - val_acc: 0.9821
Epoch 12/20
60000/60000 [==============================] - 3s - loss: 0.0240 - acc: 0.9932 - val_loss: 0.0984 - val_acc: 0.9836
Epoch 13/20
60000/60000 [==============================] - 3s - loss: 0.0230 - acc: 0.9939 - val_loss: 0.1032 - val_acc: 0.9822
Epoch 14/20
60000/60000 [==============================] - 3s - loss: 0.0236 - acc: 0.9933 - val_loss: 0.1002 - val_acc: 0.9843
Epoch 15/20
60000/60000 [==============================] - 3s - loss: 0.0184 - acc: 0.9945 - val_loss: 0.1111 - val_acc: 0.9811
Epoch 16/20
60000/60000 [==============================] - 3s - loss: 0.0201 - acc: 0.9944 - val_loss: 0.0982 - val_acc: 0.9837
Epoch 17/20
60000/60000 [==============================] - 3s - loss: 0.0186 - acc: 0.9949 - val_loss: 0.1012 - val_acc: 0.9841
Epoch 18/20
60000/60000 [==============================] - 3s - loss: 0.0179 - acc: 0.9951 - val_loss: 0.1132 - val_acc: 0.9824
Epoch 19/20
60000/60000 [==============================] - 3s - loss: 0.0189 - acc: 0.9950 - val_loss: 0.1081 - val_acc: 0.9842
Epoch 20/20
60000/60000 [==============================] - 3s - loss: 0.0168 - acc: 0.9956 - val_loss: 0.1109 - val_acc: 0.9837

# 输出训练好的模型在测试集上的表现
print('Test score:', score[0])
print('Test accuracy:', score[1])

Test score: 0.110892460335
Test accuracy: 0.9837