Selecting multiple columns from pandas DataFrame by labels and integers in single indexing

Question

I want to select columns 'd' and 'f', and the first two columns, whatever their name is, in that order, from my pandas DataFrame.

In [7]: df = pd.DataFrame({'a': [1, 2, 3], 'b': [4, 5, 6], 
   ...: 'c': [7, 8, 9], 'd': [10, 11, 12], 
   ...: 'e': [13, 14, 15], 'f': [16, 17, 18]})                                                                                                                                                                                                                                    

In [8]: df                                                                                                                                                                                                                                                                        
Out[8]: 
   a  b  c   d   e   f
0  1  4  7  10  13  16
1  2  5  8  11  14  17
2  3  6  9  12  15  18

In [9]: df[['d', 'f'] + list(df.columns[:2])]                                                                                                                                                                                                                                     
Out[9]: 
    d   f  a  b
0  10  16  1  4
1  11  17  2  5
2  12  18  3  6

Is there a better way? That is: more concise, elegant, or performant.

Answer 1

Wouldn't say this is more elegant than what you're already doing, but here are some equally concise versions of your selection:

df[['d', 'f'] + [*df][:2]]  # python >= 3.5 only

    d   f  a  b
0  10  16  1  4
1  11  17  2  5
2  12  18  3  6

This uses label based slicing. the [*df] term unpacks df 's columns into a list, then uses that to slice according to the given range. If you have multiple independent ranges to slice, either save the output of [*df] in a variable for reuse, or see below.

Performance is also hard to pin down, since these are all operations on lists/headers (what we call DataFrame metadata) instead of the actual data. So if there are any performance bottlenecks in your code, it is not here.

---

You can convert your labels into int positional indexes and index using np.r_ and df.iloc :

l = df.columns.get_loc
df.iloc[:, np.r_[l('d'), l('f'), :2]]

    d   f  a  b
0  10  16  1  4
1  11  17  2  5
2  12  18  3  6