This question is a followup from this SO question: Pandas: add columns to multiindex for any depht of index levels
but in contrast, I do have following dataframe:
index = [['A', 'B', 'C', 'D'], ['a', 'b', 'a', 'b']]
cols = [['AC', 'AC', 'BC', 'DC', 'CC'], ['ac', 'aac', 'bc', 'ac', 'bc'], ['AAc', 'AAAAc', 'BBc', 'AAc', 'BBc']]
data = np.random.random((4, 5))
df = pd.DataFrame(data=data, index=index, columns=cols)
df.columns.names = ['col_name_0', 'col_name_1', 'col_name_2']
If I apply the solution from the previous post, I do get to many columns because the level 'col_name_2' also gets broadcasted to all groups of level 0.
The solution from the cited question was:
out = df.stack(level = 1).unstack().swaplevel(1, 2, axis = 1)
But this yields:
col_name_0 AC BC CC DC
col_name_1 aac ac bc aac ac bc aac ac bc aac ac bc
col_name_2 AAAAc AAc AAAAc AAc AAAAc AAc BBc BBc BBc BBc BBc BBc AAc AAc AAc
A a 0.908180 NaN NaN 0.383903 NaN NaN NaN NaN 0.993260 NaN NaN 0.112402 NaN 0.196868 NaN
B b 0.901394 NaN NaN 0.096745 NaN NaN NaN NaN 0.260379 NaN NaN 0.723057 NaN 0.194833 NaN
The level col_name_2 are physical units which belong to the corresponding features in the level col_name_1
. Accordincly column number 1 (0 index) does not make any sense. Same for column 3. Do you know how I could i) keep the units ii) just broadcast col_name_1
across all groups?
My current approach is to drop the level col_name_2
prior to stack
and unstack
but this would require an extra dictionary to map the units to the features which is not that bad but maybe there is a more elegant solution.
How about:
df.stack(level=(1,2)).unstack(level=(-1,-2))
Output:
col_name_0 AC BC ... CC DC
col_name_2 AAAAc AAc BBc AAAAc AAc ... AAc BBc AAAAc AAc BBc
col_name_1 aac ac bc aac ac ... ac bc aac ac bc
A a 0.724763 0.688566 NaN NaN NaN ... NaN 0.854830 NaN 0.653829 NaN
B b 0.990737 0.689543 NaN NaN NaN ... NaN 0.486084 NaN 0.027718 NaN
C a 0.822234 0.122896 NaN NaN NaN ... NaN 0.580121 NaN 0.043333 NaN
D b 0.269341 0.503598 NaN NaN NaN ... NaN 0.447615 NaN 0.384507 NaN
The technical post webpages of this site follow the CC BY-SA 4.0 protocol. If you need to reprint, please indicate the site URL or the original address.Any question please contact:yoyou2525@163.com.