Is there a numpy function for sorting according to a vector with indices?

user_11599075 0
Viewed 100

I want to sort a numpy vector according to another numpy vector with indices.

import numpy as np
my_vector = np.array([9. 6. 21. 17. 12.])
my_indices = np.array([1. 4. 2. 0. 3.])

what_i_want = np.array([17. 9. 21. 12. 6.])

Is there a function for this? Something like reverse argsort

permutation numpy sorting python
3 Answers

It can be done in O(n) using advanced assignment:

my_vector = np.array([9., 6., 21., 17., 12.])
my_indices = np.array([1., 4., 2., 0., 3.])
my_result = np.empty_like(my_vector)
my_result[my_indices.astype(int)] = my_vector
my_result
# array([17.,  9., 21., 12.,  6.])

Timings on a larger example:

from timeit import timeit

a = np.random.random(1000000)
b = np.random.permutation(1000000).astype(float)

def fargsort():
    return(a[b.argsort()])

def fassign():
    c = np.empty_like(a)
    c[b.astype(int)] = a
    return c

np.all(fargsort() == fassign())
# True
timeit(fargsort,number=10)
# 1.2764090860000579
timeit(fassign,number=10)
# 0.17262099700747058
user_72073920

Use argsort as follows:

import numpy as np
my_vector = np.array([9., 6., 21., 17., 12.])
my_indices = np.array([1., 4., 2., 0., 3.])

print(my_vector[my_indices.argsort()])

Output:

[17. 9. 21. 12. 6.]

user_50523650

You can try this:

import numpy as np
my_vector = np.array([9., 6., 21., 17., 12.])
my_indices = np.array([1., 4., 2., 0., 3.])

res_sorted = [x for _,x in sorted(zip(my_indices,my_vector))]
print(res_sorted)

output:

[17.0, 9.0, 21.0, 12.0, 6.0]
user_58109500
Related