定义：

    1. model = Sequential()
    2. mode.add(Dense(input_dimension = 28*28, units = 500,))

    训练：

    1. model.compile(loss='', optimizer='', metrics=['accuracy'])
    2. model.fit(x_train, y_train, batch_size=100, epochs=20)

    预测：

    1. score = model.evaluate(x_test, y_test) # 计算正确率
    2. result = model.predict(x_test)    # 预测