熊猫：如何通过拆分从一个多索引级别向一个多索引添加级别？

Question

How can I create a new level by splitting the second level at | ?

The initial index:

MultiIndex(levels=[['A', 'B', 'C', 'D'], ['a|a_unit', 'b|b_unit', 'c|c_unit']],
       codes=[[0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3], [0, 1, 2, 0, 1, 2, 0, 1, 2, 0, 1, 2]])

Desired output:

What I tried:

# plan was to create a new column and use set_index
df.columns.to_frame().iloc[:,1].str.split('|')

EDIT: The reason why my approach did not work was the following:

Initially, I had the values in level 1 of the index separated by ' | ' to make this example simpler, I deleted the * . Without the start everything worked well, but with the start, I got an re error:

re.error: nothing to repeat at position 0

Having proper testcases is really tricky sometimes.

Answer 1

You can try with:

s=df.columns.to_frame().iloc[:,1].str.split('|')
final=(pd.DataFrame(data=df.values,columns=df.columns.get_level_values(0))
                   .T.set_index([s.str[0],s.str[1]],append=True).T)

Or:

final=(pd.DataFrame(columns=
 pd.MultiIndex.from_arrays([df.columns.get_level_values(0),s.str[0],s.str[1]])))

---

Answer 2

The answer by anky_91 is quite compact. Here is another solution which also works with this index:

MultiIndex(levels=[['A', 'B', 'C', 'D'], ['a*|*a_unit', 'b*|*b_unit', 'c*|*c_unit']],
       codes=[[0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3], [0, 1, 2, 0, 1, 2, 0, 1, 2, 0, 1, 2]])

    #  clean up the column index to have the same structure as before
    _split = [item.split('*|*') for item in df.columns.to_frame().values[:, 1]]
    _level_0 = df.columns.to_frame().values[:, 0].tolist()

    # get the old feature names (units still missing)
    idx_list = [(item[0], item[1][0], item[1][1]) for item in zip(_level_0, _split)]
    df_1.columns = pd.Index(idx_list)

I deleted the * for the sake of simplicity but doing so removed the cause why my initial approach (see anky:91's answer): df.columns.to_frame().iloc[:,1].str.split('|') did not work

Answer 3

Another method is to access your levels with index.get_level_values and split them into three indices:

idx1 = [idx.split('|')[0] for idx in df.index.get_level_values(1)]
idx2 = [idx.split('|')[1] for idx in df.index.get_level_values(1)]
df.index = [df.index.get_level_values(0), idx1, idx2]

Output

Empty DataFrame
Columns: []
Index: [(A, a, a_unit), (A, b, b_unit), (A, c, c_unit), (B, a, a_unit), (B, b, b_unit), (B, c, c_unit), (C, a, a_unit), (C, b, b_unit), (C, c, c_unit), (D, a, a_unit), (D, b, b_unit), (D, c, c_unit)]