Create columns
fst_imp: return column name in which value is index of the min value of each row.
snd_imp: value is column name in which value is index of the second small value of each row.
trd_imp: value is column name in which value is index of the third small value of each row.
A B C fst_imp snd_imp trd_imp
0 1 2 3 A B C
1 6 5 4 C B A
2 7 9 8 A C B
Here is one potential solution using numpy.argsort, the pandas.DataFrame constructor and DataFrame.join:
# Setup
import numpy as np
df = pd.DataFrame({'A': {0: 1, 1: 6, 2: 7}, 'B': {0: 2, 1: 5, 2: 9}, 'C': {0: 3, 1: 4, 2: 8}})
df.join(pd.DataFrame([df.columns.values[x] for x in np.argsort(df.values)],
columns=['fst_imp', 'snd_imp', 'trd_imp']))
[out]
A B C fst_imp snd_imp trd_imp
0 1 2 3 A B C
1 6 5 4 C B A
2 7 9 8 A C B
Or a bit more scalable...
df.join(pd.DataFrame([df.columns.values[x] for x in np.argsort(df.values)]))
[out]
A B C 0 1 2
0 1 2 3 A B C
1 6 5 4 C B A
2 7 9 8 A C B
columns=[...] arg in that case, should work ok, you may just need a different way of renaming those column headings