import numpy as np

from keras.datasets import mnist

from keras.utils import np_utils

from keras.models import Sequential,load_model

from keras.layers import *

from keras.optimizers import SGD

import os

import tensorflow as tf

# 载入数据

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

# 预处理

# 将（60000，28，28）转化为（600000，784），好输入展开层

x_train = x_train.reshape(x_train.shape[0],-1)/255.0

x_test= x_test.reshape(x_test.shape[0],-1)/255.0

# 将输出转化为one_hot编码

y_train = np_utils.to_categorical(y_train,num_classes=10)

y_test = np_utils.to_categorical(y_test,num_classes=10)

# 加载模型

if os.path.exists('model.h5'):

    print('--------load model-----------')

    model = load_model('model.h5')

else:

    # 创建网络

    model = Sequential([

        # 输入784输出10个

        Dense(units=10,input_dim=784,bias_initializer='one',activation='softmax')

    ])

    # 编译

    # 自定义优化器

    sgd = SGD(lr=0.1)

    model.compile(optimizer=sgd,

                  loss='mse',

                  # 得到训练过程中的准确率

                  metrics=['accuracy'])

    model.fit(x_train,y_train,batch_size=32,epochs=10,validation_split=0.2)

    # 保存模型

    model.save('model.h5')

# 评估模型

loss,acc = model.evaluate(x_test,y_test,)

print('\ntest loss',loss)

print('test acc',acc)

# 保存参数

model.save_weights('my_model_weights.h5')

model.load_weights('my_model_weights.h5')

# 保存模型结构

from keras.models import  model_from_json

json_string = model.to_json()

model = model_from_json(json_string)

print(json_string)