Sort numpy 2d array by multiple columns

Question

I have a 2D numpy array that looks like this

array([[5, 0],
       [3, 1],
       [7, 0],
       [2, 1]])

I'd like to (sub) sort by each column (say right to left) to get this:

array([[5, 0],
       [7, 0],
       [2, 1],
       [3, 1]])

How can I do that in numpy?

Did you try to change the column index like `array[np.argsort(array[:, 0])]`? — Ugurcan, Feb 15 '21 at 19:16
Does this answer your question? [Sorting arrays in NumPy by column](https://stackoverflow.com/questions/2828059/sorting-arrays-in-numpy-by-column) — Ugurcan, Feb 15 '21 at 19:20

anon01 · Accepted Answer · 2021-02-19T18:28:44.337

1

Numpy includes a native function for sub-sorting by columns, lexsort:

idx = np.lexsort((arr[:,0], arr[:,1]))
arr_sorted = arr[idx]

Alternatively, you can use pandas syntax if you're more familiar; this will have some memory/time overhead but should be small for < 1m rows:

arr = [
    [5,  0],
    [3,  1],
    [7,  0],
    [2,  1]
]
df = pd.DataFrame(data=arr).sort_values([1,0])
arr_sorted = df.to_numpy()

output (both):

array([[5, 0],
       [7, 0],
       [2, 1],
       [3, 1]])

edited Feb 19 '21 at 18:28

answered Feb 15 '21 at 19:42

anon01

Ah yes, but it is a very large array unfortunately – Floris Fancypants Feb 15 '21 at 19:51
@FlorisFancypants how large is very large? This should be acceptable up to a few million rows anyway – anon01 Feb 15 '21 at 19:52
oh, its like 10000, but the lexsort works too, thanks <3 – Floris Fancypants Feb 15 '21 at 20:34

S3DEV · Answer 2 · 2021-02-15T21:26:18.993

0

You can use np.lexsort to sort an array on multiple columns:

idx = np.lexsort((a[:,0], a[:,1]))

a[idx]

Output:

array([[5, 0], 
       [7, 0],
       [2, 1],
       [3, 1]])

edited Feb 15 '21 at 21:26

answered Feb 15 '21 at 19:22

S3DEV

2 Answers2