Pandas mehrstufige Gruppierungsmethode zur Sortierung

In [70]: df = pd.DataFrame(((random.randint(2012, 2016), random.choice(['tech', 'art', 'office']), '%dk-%dk'%(random.randint(2,10), random.randint(10, 20)), '') for _ in xrange(10000)), columns=['publish_time', 'classf', 'salary', 'title'])

In [71]: df.head()
Out[71]:
 publish_time classf salary title
0   2012  art 2k-19k
1   2014 office 5k-17k
2   2013 office 2k-10k
3   2013  art 5k-14k
4   2013  art 2k-14k

In [72]: df.groupby(['publish_time', 'classf', 'salary']).count()['title'].groupby(level=0, group_keys=False).nlargest(10)
Out[72]:
publish_time classf salary
2012   art  7k-13k  18
      4k-13k  16
    tech 3k-12k  14
    art  6k-16k  13
      8k-15k  13
    office 5k-18k  13
    tech 4k-14k  13