• 决策树——非正常企业数目预测


    核心步骤:

    ①将nsrxx.csv和zzsfp.csv两个表进行了合并和缩减,最终保留下了6个字段存入business.csv

    字段含义:

      id:企业id

      xf_count:作为销方次数

      gf_count:作为购方次数

      del_count:销方购方次数之差

      zfcs:作废次数

      problem:是否为问题企业

     ②导入相关扩展包

    from sklearn.model_selection import train_test_split  # 划分数据集
    from sklearn.feature_extraction import DictVectorizer  #字典特征值提取
    from sklearn.tree import DecisionTreeClassifier  # 决策树
    from sklearn.tree import export_graphviz  # 决策树可视化
    import pandas as pd

    ③获取数据

    business=pd.read_csv("./business.csv")

    ④筛选特征值和目标值

    x=business[["xf_count","gf_count","del_count","zfcs"]]        #特征值
    y=business["problem"]                   #目标值

    ⑤将特征值转化为字典格式

    x=x.to_dict(orient="records")

    ⑥划分数据集

    x_train,x_test,y_train,y_test=train_test_split(x,y)

    ⑦字典特征抽取

    transfer=DictVectorizer()
    x_train=transfer.fit_transform(x_train)
    x_test=transfer.transform(x_test)

    ⑧决策树预估器(estimator)

    estimator = DecisionTreeClassifier(criterion="entropy")  # criterion默认为'gini'系数,也可选择信息增益熵'entropy'
    estimator.fit(x_train, y_train)  # 调用fit()方法进行训练,()内为训练集的特征值与目标值

    ⑨模型评估

    • 直接对比真实值和预测值
    y_predict = estimator.predict(x_test)  # 传入测试集特征值,预测所给测试集的目标值
    print("y_predict:
    ", y_predict)
    • 计算有问题的企业数
        flag=0;
        for i in y_train:
            if(i==1):
                flag=flag+1
        for i in y_test:
            if(i==1):
                flag=flag+1
        print("有问题的企业数为:",flag)
    • 对比真实值和预测值:
    print("直接对比真实值和预测值:
    ", y_test == y_predict)
    • 计算准确率
    score = estimator.score(x_test, y_test)  # 传入测试集的特征值和目标值
    print("准确率为:
    ", score)

    ⑩决策树可视化

    export_graphviz(estimator, out_file="tree_business.dot", feature_names=transfer.get_feature_names())

    核心代码:

    def tree_business():
        # 1.获取数据
        business=pd.read_csv("./business.csv")
    
        # 2.筛选特征值和目标值
        x=business[["xf_count","gf_count","del_count","zfcs"]]        #特征值
        y=business["problem"]                   #目标值
    
        # 3.数据处理(缺失值处理,特征值——>字典类型)
            #转换为字典
        x=x.to_dict(orient="records")
    
        # 4.划分数据集
        x_train,x_test,y_train,y_test=train_test_split(x,y)
    
        # 5.字典特征抽取
        transfer=DictVectorizer()
        x_train=transfer.fit_transform(x_train)
        x_test=transfer.transform(x_test)
    
        # 6.决策树预估器(estimator)
        estimator = DecisionTreeClassifier(criterion="entropy")  # criterion默认为'gini'系数,也可选择信息增益熵'entropy'
        estimator.fit(x_train, y_train)  # 调用fit()方法进行训练,()内为训练集的特征值与目标值
    
        # 7.模型评估
        # 方法一:直接对比真实值和预测值
        y_predict = estimator.predict(x_test)  # 传入测试集特征值,预测所给测试集的目标值
        print("y_predict:
    ", y_predict)
        #计算有问题的企业数:
        flag=0;
        for i in y_train:
            if(i==1):
                flag=flag+1
        for i in y_test:
            if(i==1):
                flag=flag+1
        print("有问题的企业数为:",flag)
        print("直接对比真实值和预测值:
    ", y_test == y_predict)
        # 方法二:计算准确率
        score = estimator.score(x_test, y_test)  # 传入测试集的特征值和目标值
        print("准确率为:
    ", score)
    
        # 8.决策树可视化
        export_graphviz(estimator, out_file="tree_business.dot", feature_names=transfer.get_feature_names())
    
    
        return None

    运行结果:

     决策树可视化:

    (因图规模过大无法展示完整)

  • 相关阅读:
    IDEA中实现接口时注解@Override报错的解决方法
    从源码分析:分析Java中的StringBuilder
    人工智能,机器学习 书籍推荐
    完美解决 scipy.misc.imread 报错 TypeError: Image data cannot be converted to float
    Python这些位运算的妙用,绝对让你大开眼界
    ubuntu vmware kernel module updater
    unzip:unzip解压文件到指定目录
    设置python 虚拟环境 virtualenv django 虚拟环境
    ubuntu sublime text3 python 配置 sublime text3 python 配置
    Linux下切换python2和python3
  • 原文地址:https://www.cnblogs.com/zyj3955/p/15437833.html
Copyright © 2020-2023  润新知