在 python 中使用线性回归估算缺失值

Question

我正在尝试使用线性回归来估算 pandas 数据框中的缺失值

`

for index in [missing_data_df.horsepower.index]:
    i = 0
    if pd.isnull(missing_data_df.horsepower[index[i]]):
            #linear regression equation
            a = 0.25743277 * missing_data_df.displacement[index[i]] + 0.00958711 * 
            missing_data_df.weight[index[i]] + 25.874947903262651
            # replacing "nan" values in dataframe using .set_value
            missing_data_df.set_value(index[i],"horsepower",a) 
    i+=1

`
它正在执行。但是数据框中的缺失值（nan）没有被变量“a”中的线性回归的预测值替换。有什么建议吗？

下面是包含缺失数据的数据框 `

   >>> missing_data_df:
       mpg cylinders  displacement  horsepower  weight  acceleration  \
10    NaN       4.0         133.0       115.0  3090.0          17.5   
11    NaN       8.0         350.0       165.0  4142.0          11.5   
12    NaN       8.0         351.0       153.0  4034.0          11.0   
13    NaN       8.0         383.0       175.0  4166.0          10.5   
14    NaN       8.0         360.0       175.0  3850.0          11.0   
17    NaN       8.0         302.0       140.0  3353.0           8.0   
38   25.0       4.0          98.0         NaN  2046.0          19.0   
39    NaN       4.0          97.0        48.0  1978.0          20.0   
133  21.0       6.0         200.0         NaN  2875.0          17.0   
337  40.9       4.0          85.0         NaN  1835.0          17.3   
343  23.6       4.0         140.0         NaN  2905.0          14.3   
361  34.5       4.0         100.0         NaN  2320.0          15.8   
367   NaN       4.0         121.0       110.0  2800.0          15.4   
382  23.0       4.0         151.0         NaN  3035.0          20.5   

       model_year origin                          car_name  
10        70.0    2.0              citroen ds-21 pallas  
11        70.0    1.0  chevrolet chevelle concours (sw)  
12        70.0    1.0                  ford torino (sw)  
13        70.0    1.0           plymouth satellite (sw)  
14        70.0    1.0                amc rebel sst (sw)  
17        70.0    1.0             ford mustang boss 302  
38        71.0    1.0                        ford pinto  
39        71.0    2.0       volkswagen super beetle 117  
133       74.0    1.0                     ford maverick  
337       80.0    2.0              renault lecar deluxe  
343       80.0    1.0                ford mustang cobra  
361       81.0    2.0                       renault 18i  
367       81.0    2.0                         saab 900s  
382       82.0    1.0                    amc concord dl

`

Answer 1

您可以为此使用 apply 和 lambda：

missing_data_df['horsepower']= missing_data_df.apply(
    lambda row: 
            0.25743277 * row.displacement + 0.00958711 * row.weight + 25.874947903262651 
            if np.isnan(row.horsepower) else row.horsepower, axis=1)

Answer 2

几件事

1. missing_data_df.horsepower 没有缺失值
2. missing_data_df.weight，您公式中的一个变量，确实有缺失值
3. 如果 hp = 0.25743277 * disp + 0.00958711 * weight + 25.874947903262651
   那么重量 = (0.25743277 * disp + 25.874947903262651 - hp) / -0.00958711

计算体重试试

for idx in missing_data_df.index:
    if pd.isnull(missing_data_df.loc[idx,"weight"]):
        disp = missing_data_df.loc[idx,"displacement"]
        hp = missing_data_df.loc[idx,"horsepower"]
        missing_data_df.loc[idx,"weight"] = (0.25743277 * disp + 25.874947903262651 - hp) / -0.00958711

一般来说，.loc[] 和 .iloc[] 是查找或设置值时更好的方法