填充NaN值函数异常：原有Age值被全部置为NaN问题排查

核心问题 #

你的agemaking函数只在Age为NaN的情况下返回了填充值，但对于非NaN的Age，函数没有任何返回语句——Python中无return的函数默认返回None，而pandas会把None解析为NaN，这就是原有非缺失值全部变成NaN的根源。

修复步骤 #

1. 修正agemaking逻辑，确保所有分支都有返回值：将else: return Age移到外层if的外面，保证非NaN的Age直接返回原值。
2. 避免操作全局变量train_titanic，让preprocessing函数处理传入的df参数，保证函数封装性。
3. 可选优化：用groupby+transform替代apply，代码更简洁且效率更高。

修复后的代码 #

方案1：修正原apply逻辑

import pandas as pd

def preprocessing(df):
    # 计算各分组的中位数
    median_male_3 = df[(df["Sex"] == "male") & (df["Pclass"] == 3)]["Age"].median()
    median_male_2 = df[(df["Sex"] == "male") & (df["Pclass"] == 2)]["Age"].median()
    median_male_1 = df[(df["Sex"] == "male") & (df["Pclass"] == 1)]["Age"].median()
    median_female_3 = df[(df["Sex"] == "female") & (df["Pclass"] == 3)]["Age"].median()
    median_female_2 = df[(df["Sex"] == "female") & (df["Pclass"] == 2)]["Age"].median()
    median_female_1 = df[(df["Sex"] == "female") & (df["Pclass"] == 1)]["Age"].median()

    def agemaking(para):
        Age = para[0]
        bookclass = para[1]
        sex = para[2]
        
        if pd.isnull(Age):
            if bookclass == 3 and sex == "male":
                return median_male_3
            elif bookclass == 2 and sex == "male":
                return median_male_2
            elif bookclass == 1 and sex == "male":
                return median_male_1
            elif bookclass == 3 and sex == "female":
                return median_female_3
            elif bookclass == 2 and sex == "female":
                return median_female_2
            elif bookclass == 1 and sex == "female":
                return median_female_1
            else:
                return Age
        # 非NaN情况直接返回原值
        return Age
    
    # 处理传入的df，而非全局变量
    df['Age'] = df[['Age','Pclass','Sex']].apply(agemaking, axis=1)
    return df

# 调用示例
# train_titanic = preprocessing(train_titanic)

方案2：更高效的groupby+transform实现

import pandas as pd

def preprocessing(df):
    # 按Sex和Pclass分组，自动用每组中位数填充Age缺失值
    df['Age'] = df.groupby(['Sex', 'Pclass'])['Age'].transform(
        lambda x: x.fillna(x.median())
    )
    return df

# 调用示例
# train_titanic = preprocessing(train_titanic)

该方案无需手动计算每个分组的中位数，pandas会自动完成分组与填充，代码更简洁，处理大数据集时效率也更高。

内容的提问来源于stack exchange，提问作者Simon