np.lexsort switch between ascending and descending order

user_6212718 0
Viewed 412

I have an numpy array with x columns and want to sort by multiple columns (some of which may be of type np.str_. I know that I can do this using np.lexsort.

Is there a way to specify if ascending / descending order for each sorting column?

Example: I know that I can sort via multiple columns as follows: (EDITED TO INDICATE STRING COLUMNS!

import numpy as np
arr = np.array([list("aaabbb"),[1,2,3,1,4,3],[1,2,3,4,6,6]]).T # Define arr
idx = np.lexsort([arr[:,1], arr[:,2]]) # sort by column 2 and then by column 1 (i.e. reversed order)
arr = arr[idx]

I also understand that I can sort in descending order as follows:

arr = arr[idx[::-1]]

This results in first sort column (column 2) to be descending and the subsequent columns in ascending order.

But how do I specify that I want first sort (column 2) in descending order and subsequent sort (column 1) descending order so that I get the following.

desired OUTPUT:

array([['b', 4, 6],
       ['b', 3, 6],
       ['b', 1, 4],
       ['a', 3, 3],
       ['a', 2, 2],
       ['a', 1, 1]]

So basically for my example I am looking something equivalent to:

df = pd.DataFrame(arr, columns=list("abc"))
df.sort_values(by=["c","b"], ascending=[False, False])

In general I want to be able to specifcy (i) columns to be sorted and (ii) sorting order per column (ascending/descending).

numpy sorting python
1 Answers

Try using the minus sign for reversing, it basically sorts by the negative of each value, which actually works:

idx = np.lexsort([arr[:,1], -arr[:,2]])

Output:

array([[2, 3, 6],
       [2, 4, 6],
       [2, 1, 4],
       [1, 3, 3],
       [1, 2, 2],
       [1, 1, 1]])

If there would be strings, try:

idx = np.lexsort([arr[:,1], arr[:,2]])
arr = arr[idx]
arr[:, 0] = arr[:, 0][::-1]

Output:

array([['2', '1', '1'],
       ['2', '2', '2'],
       ['2', '3', '3'],
       ['1', '1', '4'],
       ['1', '3', '6'],
       ['1', '4', '6']], 
      dtype='<U11')

Edit:

With the new edit:

arr = arr[::-1]

Would work:

array([['b', '4', '6'],
       ['b', '3', '6'],
       ['b', '1', '4'],
       ['a', '3', '3'],
       ['a', '2', '2'],
       ['a', '1', '1']], 
      dtype='<U1')
user_87083640
Related