Okay, so, I have a 4x2 numpy ndarray, and I want to sort it lexicographically. That is, if I have the array
[[0,0],
[1,1],
[0,1],
[1,0]]
I want it to become
[[0,0],
[0,1],
[1,0],
[1,1]]
How do I do this?
Asked
Active
Viewed 1,515 times
3
Pedro Carvalho
- 565
- 1
- 6
- 26
-
1`l.sort()` does the trick for normal lists – Bhargav Rao Jan 21 '15 at 18:10
-
Doesn't seem to be doing it. Going l.sort() turns [[2,1], [0,2]] into [[1,2], [0,2]] instead of [[0,2], [2,1]] (when it's an ndarray). – Pedro Carvalho Jan 21 '15 at 18:27
-
`l.sort()` converts `l = [[2,1], [0,2]]` into `[[0, 2], [2, 1]]` only for a normal list and not for an numpy array – Bhargav Rao Jan 21 '15 at 18:31
1 Answers
5
You can use numpy's lexsort. Lexsort, though, sorts using the last column as the primary key. One way to get what you want is to specify the columns explicitly:
x[np.lexsort((x[:,1], x[:,0]))]
# array([[0, 0],
# [0, 1],
# [1, 0],
# [1, 1]])
tom10
- 67,082
- 10
- 127
- 137
-
-
@PedroCarvalho: For an h-by-w array, the worst case performance is `O(h*w*log(h))`, since it's a comparison sort on an input of length h where comparisons can take worst-case O(w) time. If rows usually differ in the first few columns, the expected performance will be `O(h*log(h))`, since comparisons will take expected O(1) time. – user2357112 May 25 '15 at 23:54
-
Also, rather than listing columns individually, you can do `x[np.lexsort(x.T[::-1])]`. (That's still way too many intermediate steps to just do a lexicographic sort, though. NumPy's sorting API seems to be put together in a really weird way.) – user2357112 May 25 '15 at 23:57