python - 用其他列值填充 NaN 列，复制新行

Question

我有一些令人困惑的操作，试图使用以下一般形式在数据集上有效地完成:

id,date,ind_1,ind_2,ind_3,ind_4
1,2014-01-01,ind_1,NaN,NaN,NaN
2,2014-01-02,ind_1,NaN,ind_3,NaN
3,2014-01-03,ind_1,ind_2,ind_3,NaN

我试图弄清楚如何创建一个新列“ind_all”，其中填充任何非空“ind”列。这很简单。我可以使用.idxmax()。然而，棘手的部分是我每行可以有多个“ind”。这意味着当存在重复项时我需要创建一条新记录。上面的例子最终应该看起来像这样:

id,date,ind_1,ind_2,ind_3,ind_4,ind_all
1,2014-01-01,ind_1,NaN,NaN,NaN,ind_1
2,2014-01-02,ind_1,NaN,ind_3,NaN,ind_1
2,2014-01-02,ind_1,NaN,ind_3,NaN,ind_3
3,2014-01-03,ind_1,ind_2,ind_3,NaN,ind_1
3,2014-01-03,ind_1,ind_2,ind_3,NaN,ind_2
3,2014-01-03,ind_1,ind_2,ind_3,NaN,ind_3

一如既往，我们非常感谢任何提示或技巧!

Answer 1

有一个基于merge的解决方案，使用melt/stack来构建RHS。

v = (df.drop('date', 1)
       .melt('id')
       .drop('variable', 1)
       .dropna()
       .rename({'value' : 'ind_all'}, axis=1)
)

df.merge(v)

   id        date  ind_1  ind_2  ind_3  ind_4 ind_all
0   1  2014-01-01  ind_1    NaN    NaN    NaN   ind_1
1   2  2014-01-02  ind_1    NaN  ind_3    NaN   ind_1
2   2  2014-01-02  ind_1    NaN  ind_3    NaN   ind_3
3   3  2014-01-03  ind_1  ind_2  ind_3    NaN   ind_1
4   3  2014-01-03  ind_1  ind_2  ind_3    NaN   ind_2
5   3  2014-01-03  ind_1  ind_2  ind_3    NaN   ind_3

或者，

df.merge(df.drop('date', 1)
           .set_index('id')
           .stack()
           .reset_index(1, drop=True)
           .to_frame('ind_all'), 
         left_on='id', 
         right_index=True
)

   id        date  ind_1  ind_2  ind_3  ind_4 ind_all
0   1  2014-01-01  ind_1    NaN    NaN    NaN   ind_1
1   2  2014-01-02  ind_1    NaN  ind_3    NaN   ind_1
1   2  2014-01-02  ind_1    NaN  ind_3    NaN   ind_3
2   3  2014-01-03  ind_1  ind_2  ind_3    NaN   ind_1
2   3  2014-01-03  ind_1  ind_2  ind_3    NaN   ind_2
2   3  2014-01-03  ind_1  ind_2  ind_3    NaN   ind_3