Why is numpy.select slower than pandas.Dataframe.apply?

Question

Say, I have the following dataframe:

df = pd.DataFrame({\'a\':[\'a\',\'b\',\'c (not a)\', \'this is (smiley)\']})

I want to produce the follo