I have found this to be really useful:
df = pd.DataFrame({'A' : range(0,10) * 2, 'B' : np.random.randint(20,30,20)})
# A ascending, B descending
df.sort(**skw(columns=['A','-B']))
# A descending, B ascending
df.sort(**skw(columns=['-A','+B']))
Note that unlike the standard columns=,ascending= arguments, here column names and their sort order are in the same place. As a result your code gets a lot easier to read and maintain.
Note the actual call to .sort is unchanged, skw (sortkwargs) is just a small helper function that parses the columns and returns the usual columns= and ascending= parameters for you. Pass it any other sort kwargs as you usually would. Copy/paste the following code into e.g. your local utils.py then forget about it and just use it as above.
# utils.py (or anywhere else convenient to import)
def skw(columns=None, **kwargs):
""" get sort kwargs by parsing sort order given in column name """
# set default order as ascending (+)
sort_cols = ['+' + col if col[0] != '-' else col for col in columns]
# get sort kwargs
columns, ascending = zip(*[(col.replace('+', '').replace('-', ''),
False if col[0] == '-' else True)
for col in sort_cols])
kwargs.update(dict(columns=list(columns), ascending=ascending))
return kwargs