BUG: stacking with MultiIndex column with some unused level uniques fail...

[changhiskhan]

...s #3170

[hayd]

Using the example from #3170, seems to be a rogue column: :S ?

In [15]: df.unstack('ID').resample('W-THU').stack('ID')
Out[15]: 
Params        NAME  VAR1
DATE       ID           
2013-01-03 1   NaN  69.0
           2   NaN  69.0
2013-01-10 1   NaN  76.0
           2   NaN  73.5

[changhiskhan]

oops you're right. i'll put in a fix

[hayd]

One last thing, ID stack loses the column.name:

In [14]: df.unstack('ID').resample('W-THU').stack('Params')
Out[14]: 
ID                  1     2
DATE       Params          
2013-01-03 VAR1    69  69.0
2013-01-10 VAR1    76  73.5

In [15]: df.unstack('ID').resample('W-THU').stack('ID')
Out[15]: 
               VAR1
DATE       ID      
2013-01-03 1   69.0
           2   69.0
2013-01-10 1   76.0
           2   73.5

I think because:

new_columns = new_columns - drop_cols

drops the column name (i.e. 'Params' in the case of the example). Maybe save it with:

cols_name = new_columns.name
new_columns = new_columns - drop_cols
new_columns.name = cols_name

Will need to add that name to xp in the test, since assert_frame_equal tests column.name now :).

(Probably - should keep the indexes name...)

[hayd]

Also: thanks for fixing this! :)

[changhiskhan]

good catch :) thanks!

I'll fix it later today.

[hayd]

I'm going to make include these two commits in a PR with some name functionality of - (for MultiIndex and Index), and will fix the name bit. Hope that's ok :)