按年份分组并将函数应用于另一列 - Python，Pandas

Question

我有一个多索引、多列数据框：

  start  A          B             C          D
0 2019  35.156667   51.621111   18.858684    1  
1 2019  NaN         50.211905   18.991290   -1  
2 2019  42.836250   58.778235   18.788889    1  
3 2020  NaN         8.188000    17.805833    1      
4 2020  42.568000   55.907143   17.300000   -1  
5 2021  46.458333   42.293750   26.322500    1  
6 2021  43.675000   60.475000   29.520000    1

每年（“开始”列）我想在 A 列中填充 NaN，如果 D>0，则使用前向值，如果 D

  start  A          B             C          D
0 2019  35.156667   51.621111   18.858684    1  
1 2019  35.156667   50.211905   18.991290   -1  
2 2019  42.836250   58.778235   18.788889    1  
3 2020  42.568000   8.188000    17.805833    1      
4 2020  42.568000   55.907143   17.300000   -1  
5 2021  46.458333   42.293750   26.322500    1  
6 2021  43.675000   60.475000   29.520000    1

我正在尝试：

df[['A','D']] = df.groupby('start').apply(lambda x: x['A'].fillna(method='ffill') if x['D']>0 else x['A'].fillna(method='bfill'))

但我收到如下错误：

The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all().

有什么帮助吗？谢谢

Answer 1

将GroupBy.agg 用于每组的前向和后向填充，然后通过numpy.where 设置值：

df1 = df.groupby('start')['A'].agg(['ffill','bfill'])
print (df1)
       ffill      bfill
0  35.156667  35.156667
1  35.156667  42.836250
2  42.836250  42.836250
3        NaN  42.568000
4  42.568000  42.568000
5  46.458333  46.458333
6  43.675000  43.675000

print (df1.columns)
Index(['ffill', 'bfill'], dtype='object')

df['A'] = np.where(df['D'] < 0, df1['ffill'], df1['bfill'])
print (df)
   start          A          B          C  D
0   2019  35.156667  51.621111  18.858684  1
1   2019  35.156667  50.211905  18.991290 -1
2   2019  42.836250  58.778235  18.788889  1
3   2020  42.568000   8.188000  17.805833  1
4   2020  42.568000  55.907143  17.300000 -1
5   2021  46.458333  42.293750  26.322500  1
6   2021  43.675000  60.475000  29.520000  1

另外，如果每组只有 2 个值并检查 D 不是必需的（取决于数据）：

df['A'] =  df.groupby('start')['A'].apply(lambda x: x.ffill().bfill())
print (df)
   start          A          B          C  D
0   2019  35.156667  51.621111  18.858684  1
1   2019  35.156667  50.211905  18.991290 -1
2   2019  42.836250  58.778235  18.788889  1
3   2020  42.568000   8.188000  17.805833  1
4   2020  42.568000  55.907143  17.300000 -1
5   2021  46.458333  42.293750  26.322500  1
6   2021  43.675000  60.475000  29.520000  1