Sort by multiple keys using different orderings

Question

Possible Duplicate:
How to write Python sort key functions for descending values

In Python 3, it's pretty easy to sort a list of objects lexicographically using multiple keys. For example:

items.sort(key = lambda obj: obj.firstname, obj.lastname)

The reverse argument lets you specify whether you want ascending or descending order. But what do you do in the case where you want to sort by multiple keys, but you want to sort using descending order for the first key, and ascending order for the second?

For example, suppose we have an object with two attributes, points and name, where points is an int and name is a str. We want to sort a list of these objects by points in descending order (so that the object with the greatest number of points comes first), but for objects with an equal number of points, we want to sort these by name in alphabetical (ascending) order.

How can this be achieved?

Ints are a special case, in that you can just use the negative to get reverse ordering. I'd be more interested in the case where they're both strings. — Mark Ransom, Jul 13 '12 at 18:45

score 38 · Accepted Answer · answered Jul 13 '12 at 18:52

There is no built-in way to handle this. For the general case, you must sort twice: first by the secondary sort, then by the primary sort. As @Mark Ransom mentioned in his comment, in many cases the variables are numeric, and so you can use the negative value to flip the ordering.

If you know the type of the variable you're trying to sort on and how to work with it, you could also write a key function that returns a decreasing value for increasing keys. See this thread for an example for strings. (Basically, you take the negative of the ASCII numerical value of the characters.)

In Python 2, you could also use a cmp function instead of a key, but this will likely make the sort slower. Whether it will make it too slow depends on how big and unsorted the list is. In Python 3, the cmp argument is gone, but as @Mark Ransom notes you could use cmp_to_key.

Great answer, however `cmp` no longer exists in python3 (I think). — mgilson, Jul 13 '12 at 18:54
You're right, I didn't notice this question was specifically about Python 3. I edited my answer. — BrenBarn, Jul 13 '12 at 18:56

score 17 · Answer 2 · answered Jul 13 '12 at 19:47

17

items.sort(key = lambda obj: (obj.firstname, [(-ord(c) for c in obj.lastname)]))

answered Jul 13 '12 at 19:47

Marco de Wit

2,686
18
22

1

won't work on datetime... – andilabs Aug 19 '15 at 16:59
@andi the way to may be converting date unix timestampl int http://stackoverflow.com/questions/1259219/django-datefield-to-unix-timestamp – andilabs Aug 19 '15 at 17:23

Mark Ransom · Answer 3 · 2012-07-13T19:00:55.727

8

There is functools.cmp_to_key to convert a comparison function into a key compatible with the sorting functions. This was provided for sorts which used a comparison function in Python 2 and needed to be converted to Python 3 which no longer allows them.

Edit: There's also a suggestion in the Python wiki under the heading Sort Stability and Complex Sorts to do the sort in multiple passes, from least significant key to most significant. This works because Python's sort is guaranteed to be stable, so the previous order is maintained when equivalent keys are encountered.

edited Jul 13 '12 at 19:00

answered Jul 13 '12 at 18:53

Mark Ransom

299,747
42
398
622

Thank you for suggestion stable sort allows to perform complex sort in multiple iterations. Interesting is it not slower than using 2.x cmp. I believe that in most complex cases cmp would be nice to have and it's removal was an mistake. I am afraid to use cmp_to_key because who knows it might also be dropped at some point. – Dmitriy Sintsov Nov 15 '16 at 09:59

Sort by multiple keys using different orderings

3 Answers3

Linked

Related