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

#建立訓練資料和測試資料，包括訓練特徵集、訓練標籤和測試特徵集、測試標籤	
(train_feature, train_label),\
(test_feature, test_label) = mnist.load_data()

#將 Features 特徵值換為 784個 float 數字的 1 維向量
train_feature_vector =train_feature.reshape(len(train_feature), 784).astype('float32')
test_feature_vector = test_feature.reshape(len( test_feature), 784).astype('float32')

#Features 特徵值標準化
train_feature_normalize = train_feature_vector/255
test_feature_normalize = test_feature_vector/255

#label 轉換為 One-Hot Encoding 編碼
train_label_onehot = np_utils.to_categorical(train_label)
test_label_onehot = np_utils.to_categorical(test_label)

#建立模型
model = Sequential()
#輸入層：784, 隱藏層：256，輸出層：10
model.add(Dense(units=256, 
                input_dim=784, 
                kernel_initializer='normal', 
                activation='relu'))
model.add(Dense(units=10, 
                kernel_initializer='normal', 
                activation='softmax'))
#定義訓練方式
model.compile(loss='categorical_crossentropy', 
              optimizer='adam', metrics=['accuracy'])

#以(train_feature_normalize,train_label_onehot)資料訓練，
#訓練資料保留 20% 作驗證,訓練10次、每批次讀取200筆資料，顯示簡易訓練過程
train_history =model.fit(x=train_feature_normalize,
                         y=train_label_onehot,validation_split=0.2, 
                         epochs=10, batch_size=200,verbose=2)

#評估準確率
scores = model.evaluate(test_feature_normalize, test_label_onehot)
print('\n準確率=',scores[1])
   
#將模型儲存至 HDF5檔案中
model.save('Mnist_mlp_model.h5')
print("Mnist_mlp_model.h5 模型儲存完畢!")
del model