Pandas实战：统计目标变量取特定值时属性的出现次数

现有如下Pandas数据集df：

Age     Income  Student     Credit Rating     Loan
0   <=30    high    no          fair              no
1   <=30    high    no          excellent         no
2   31-40   high    no          fair              yes
3   >40     medium  no          fair              yes
4   >40     low     yes         excellent         no
5   31-40   low     yes         excellent         yes

已定义属性列表：

attributes = ["Age", "Income", "Student", "Credit Rating"]

以及属性值字典：

attribute_values = {
    "Age": ["<=30", "31-40", ">40"],
    "Income": ["low", "medium", "high"],
    "Student": ["yes", "no"],
    "Credit Rating": ["fair", "excellent"]
}

目前已统计出各属性值的出现次数：

attribute Age
<=30 2
31-40 2
>40 2
attribute Income
low 2
medium 1
high 3
attribute Student
yes 4
no 2
attribute Credit Rating
fair 3
excellent 3

需要进一步统计每个属性值对应的目标变量Loan为yes或no的次数，例如Age<=30的2条数据中Loan均为no，Credit Rating为fair的3条数据中1条Loan为no、2条为yes。

方法1：使用groupby + value_counts #

直接对每个属性分组后，统计Loan的取值频次，自动补全缺失结果为0，输出规整：

import pandas as pd

for attr in attributes:
    print(f"attribute {attr}")
    # 分组后统计Loan的频次，unstack将结果转为宽表，fill_value补0
    print(df.groupby(attr)["Loan"].value_counts().unstack(fill_value=0))
    print()

输出示例（以Age为例）：

attribute Age
Loan      no  yes
Age              
<=30       2    0
31-40      0    2
>40        1    1

方法2：使用pd.crosstab #

交叉表可以直接生成属性与目标变量的频次统计，代码更简洁：

import pandas as pd

for attr in attributes:
    print(f"attribute {attr}")
    print(pd.crosstab(df[attr], df["Loan"]))
    print()

输出格式与方法1完全一致，适合快速生成统计表格。

方法3：结合预定义属性值遍历统计 #

如果需要严格按照attribute_values中定义的顺序输出，可手动遍历筛选统计：

for attr in attributes:
    print(f"attribute {attr}")
    for val in attribute_values[attr]:
        # 筛选当前属性值的子集
        subset = df[df[attr] == val]["Loan"]
        # 统计no和yes的次数
        no_count = (subset == "no").sum()
        yes_count = (subset == "yes").sum()
        print(f"{val} no:{no_count} yes:{yes_count}")
    print()

输出示例：

attribute Age
<=30 no:2 yes:0
31-40 no:0 yes:2
>40 no:1 yes:1

内容的提问来源于stack exchange，提问作者LeGOATJames23