wide & deep 模型

我们可以把一些比较常见的特征直接把他记录下来作为wide模型，把一些需要泛化的特征让他进行深度学习，形成deep模型，所以显而易见，wide模型使用多类型的叉乘特征变换，记忆特定的特征组合，但它的限制就是难以归纳以前没出现过的组合，那这就需要人工提供特征。而deep模型虽然较为复杂，但是可以提高整体模型的泛化水平，这也就是线性模型+DNN的组合。

import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt
import numpy as np
import random
import pandas as pd
import os 
import sys
#导入数据集
from sklearn.datasets import fecth_california_housing
housing = fectch_california_housing()
#用sklearn进行数据集的划分
from sklearn.model_selection import train_test_split
x_train_all,x_test,y_train_all,y_test = train_test_split(	
	housing.data,housing.target.random.state = 7
)
x_train,x_vaild,y_train,y_vaild = train_test_split(
	x_train_all,y_train_all,random.state = 11 ,test_size = 0.2
)

#对数据集归一化
from sklearn.preprocessing import StandardScaler
#创建一个类
scaler = StandardScaler()
x_train_scaler = scaler.fit_transform(x_train)
y_vaild_scaler = scaler.transform(x_vaild)
x_test_scaler = scaler.transform(x_test)

#搭建模型
#deep model
input = keras.layers.Input (shape = x_train.shape[1:])
hidden1 = keras.layers.Dense(30,activation = 'relu')(input)
hidden2 = keras.layers.Dense(30,activation = 'relu')(hidden1)
#将deep模型的输出和wide模型的输入拼接起来（假设deep和wide的input相同）
concat = keras.layers.concatenate([input,hidden2])
output = keras.layers.Dense(1)(concat)

model = keras.models.Model(inputs = [input],
			   outputs = [output])
#编译并且训练模型
model.compile(loss = 'mse' , optimizer = 'adam')
model.fit(x_train_scaler,y_train,
	  validation_data =(x_valid_scaler,y_valid),
	  epochs = 10)
mode1.evaluate(x_test_scaler,y_test)