使用pmml跨平台部署机器学习模型Demo——房价预测

基于房价数据，在python中训练得到一个线性回归的模型，在JavaWeb中加载模型完成房价预测的功能。

一、训练、保存模型

工具：PyCharm-2017、Python-39、sklearn2pmml-0.76.1。

1.训练数据house_price.csv

No	square_feet	price
1	150	6450
2	200	7450
3	250	8450
4	300	9450
5	350	11450
6	400	15450
7	600	18450

2.训练、保存模型

import sklearn2pmml as pmml

from sklearn2pmml import PMMLPipeline

from sklearn import linear_model as lm

import os

import pandas as pd

def save_model(data, model_path):

    pipeline = PMMLPipeline([("regression", lm.LinearRegression())])

    pipeline.fit(data[["square_feet"]], data["price"])

    pmml.sklearn2pmml(pipeline, model_path, with_repr=True)

if __name__ == "__main__":

    data = pd.read_csv("house_price.csv")

    model_path = model_path = os.path.dirname(os.path.abspath(__file__)) + "/my_liner_model.pmml"

    save_model(data, model_path)

    print("模型保存完成。")

二、JavaWeb应用开发

工具：IntelliJ IDEA-2018、jdk-14.0.2、Tomcat-9.0.37。

创建maven项目，加入依赖项

    <dependencies>

        <dependency>

            <groupId>org.jpmml</groupId>

            <artifactId>pmml-evaluator</artifactId>

            <version>1.4.15</version>

        </dependency>

        <dependency>

            <groupId>com.sun.xml.bind</groupId>

            <artifactId>jaxb-core</artifactId>

            <version>2.2.11</version>

        </dependency>

        <dependency>

            <groupId>javax.xml</groupId>

            <artifactId>jaxb-api</artifactId>

            <version>2.1</version>

        </dependency>

        <dependency>

            <groupId>com.sun.xml.bind</groupId>

            <artifactId>jaxb-impl</artifactId>

            <version>2.2.11</version>

        </dependency>

        <dependency>

            <groupId>javax.servlet</groupId>

            <artifactId>javax.servlet-api</artifactId>

            <version>3.0.1</version>

        </dependency>

    </dependencies>

项目结构为

使用pmml跨平台部署机器学习模型Demo——房价预测

界面——index.jsp

<%@ page contentType="text/html;charset=UTF-8" language="java" %>

<html>

<head>

    <title>使用pmml跨平台部署机器学习模型Demo</title>

</head>

<body>

<h2>使用pmml跨平台部署机器学习模型Demo——房价预测</h2>

<form name="form" method="post" action="/PredictServlet">

    <label>房子英尺数（整数）：</label>

    <input type="text" name="feet" required>

    <button type="submit">预测房价</button>

</form>

<div>

    <label>预测价格为：</label>

    ${price}

</div>

</body>

</html>

Servlet类——PredictServlet.java

package servlet;

import service.PredictService;

import service.imp.PredictServiceImp;

import javax.servlet.ServletException;

import javax.servlet.annotation.WebServlet;

import javax.servlet.http.HttpServlet;

import javax.servlet.http.HttpServletRequest;

import javax.servlet.http.HttpServletResponse;

import java.io.IOException;

@WebServlet("/PredictServlet")

public class PredictServlet extends HttpServlet {

    protected void doPost(HttpServletRequest request, HttpServletResponse response) throws ServletException, IOException {

        PredictService predictService = new PredictServiceImp();

        String feet_str = request.getParameter("feet"); //获取前端传来的值

        int feet = Integer.parseInt(feet_str);

        double price = predictService.getPredictedPrice(feet); //预测

        //请求转发，返回结果

        request.setAttribute("price", price);

        request.getRequestDispatcher("/index.jsp").forward(request, response);

    }

    protected void doGet(HttpServletRequest request, HttpServletResponse response) throws ServletException, IOException {

        this.doPost(request, response);

    }

}

Service接口——PredictService.java

package service;

public interface PredictService {

    public double getPredictedPrice(int feet);

}

Service实现类——PredictServiceImp.java

package service.imp;

import org.dmg.pmml.FieldName;

import org.dmg.pmml.PMML;

import org.jpmml.evaluator.*;

import service.PredictService;

import java.io.FileInputStream;

import java.io.InputStream;

import java.util.HashMap;

import java.util.LinkedHashMap;

import java.util.List;

import java.util.Map;

public class PredictServiceImp implements PredictService {

    public double getPredictedPrice(int feet) {

        String model_path = "D:\\my_liner_model.pmml"; //pmml模型文件存放路径

        Evaluator model = loadModel(model_path); //加载模型

        Object r = predict(model, feet); //预测

        double result = Double.parseDouble(String.format("%.2f", r)); //格式化

        return result;

    }

    private static Evaluator loadModel(String model_path){

        PMML pmml = new PMML(); //定义PMML对象

        InputStream inputStream; //定义输入流

        try {

            inputStream = new FileInputStream(model_path); //输入流接到磁盘上的模型文件

            pmml = org.jpmml.model.PMMLUtil.unmarshal(inputStream); //将输入流解析为PMML对象

        }catch (Exception e){

            e.printStackTrace();

        }

        ModelEvaluatorFactory modelEvaluatorFactory = ModelEvaluatorFactory.newInstance(); //实例化一个模型构造工厂

        Evaluator evaluator = modelEvaluatorFactory.newModelEvaluator(pmml); //将PMML对象构造为Evaluator模型对象

        return evaluator;

    }

    private static Object predict(Evaluator evaluator, int feet){

        Map<String, Integer> data = new HashMap<String, Integer>(); //定义测试数据Map，存入各元自变量

        data.put("square_feet", feet); //键"square_feet"为自变量的名称，应与训练数据中的自变量名称一致

        List<InputField> inputFieldList = evaluator.getInputFields(); //得到模型各元自变量的属性列表

        Map<FieldName, FieldValue> arguments = new LinkedHashMap<FieldName, FieldValue>();

        for (InputField inputField : inputFieldList) { //遍历各元自变量的属性列表

            FieldName inputFieldName = inputField.getName();

            Object rawValue = data.get(inputFieldName.getValue()); //取出该元变量的值

            FieldValue inputFieldValue = inputField.prepare(rawValue); //将值加入该元自变量属性中

            arguments.put(inputFieldName, inputFieldValue); //变量名和变量值的对加入LinkedHashMap

        }

        Map<FieldName, ?> result = evaluator.evaluate(arguments); //进行预测

        List<TargetField> targetFieldList = evaluator.getTargetFields(); //得到模型各元因变量的属性列表

        FieldName targetFieldName = targetFieldList.get(0).getName(); //第一元因变量名称

        Object targetFieldValue = result.get(targetFieldName); //由因变量名称得到值

        return targetFieldValue;

    }

}