How to sort a Pandas DataFrame according to multiple criteria?

Question

I have the following DataFrame containing song names, their peak chart positions and the number of weeks they spent at position no 1:

Answer 1

df.sort(['Peak', 'Weeks'], ascending=[True, False], inplace=True)

If you want the sorted result for future use, inplace=True is required.

Answer 2

On pandas 0.9.1 and higher this should work (this is with 0.10.0b1):

(Edit: As of Pandas 0.19, method sort_index is deprecated. Prefer sort_values)

In [23]: songs.sort_index(by=['Peak', 'Weeks'], ascending=[True, False])
Out[23]: 
                                      Song  Peak  Weeks
10                           She Loves You     1     36
118                               Hey Jude     1     27
20                I Want To Hold Your Hand     1     24
22                       Can't Buy Me Love     1     17
56                                   Help!     1     17
76                        Paperback Writer     1     16
109                   All You Need Is Love     1     16
45                             I Feel Fine     1     15
29                      A Hard Day's Night     1     14
48                          Ticket To Ride     1     14
85                           Eleanor Rigby     1     14
87                        Yellow Submarine     1     14
173            The Ballad Of John And Yoko     1     13
60                             Day Tripper     1     12
61                      We Can Work It Out     1     12
117                           Lady Madonna     1      9
8                           From Me To You     1      7
115                          Hello Goodbye     1      7
155                               Get Back     1      6
2                         Please Please Me     2     20
107                   Magical Mystery Tour     2     16
176                              Let It Be     2     14
93                              Penny Lane     2     13
92               Strawberry Fields Forever     2     12
0                               Love Me Do     4     26
166                          Come Together     4     10
157                              Something     4      9
58                               Yesterday     8     21
135                   Back In The U.S.S.R.    19      3
164                     Here Comes The Sun    58     19
96   Sgt. Pepper's Lonely Hearts Club Band    63     12
105     With A Little Help From My Friends    63      7

Answer 3

In case, if the dtypes of 'Peak' and 'Week' are not 'int' or 'float', then use the following command.

df.convert_objects(convert_numeric=True).sort_values(['Peak', 'Weeks'], ascending=[True, False], inplace=True)

Answer 4

Since pandas 0.17.0, sort is deprecated and replaced by sort_values:

df.sort_values(['Peak', 'Weeks'], ascending=[True, False], inplace=True)

Answer 5

By using .sort()

df.sort(['Peak', 'Weeks'], ascending=[True, False])

Will sort into ascending order of peak position, then within that descending order of length in charts.