How to down sample an array in python without a for loop

Question

Is there a 'pythonic' way to cleanly down-sample without multiple for loops?

This example below is the type of for loop I wish to get rid of.

Minimum working example:

import numpy as np
unsampled_array = [1,3,5,7,9,11,13,15,17,19]
number_of_samples = 7
downsampled_array = []
downsampling_indices = np.linspace(0, len(unsampled_array)-1, number_of_samples).round()
for index in downsampling_indices:
    downsampled_array.append(unsampled_array[int(index)])
print(downsampled_array)

Result:

>>> [ 1  5  7  9 13 17 19]

Noam Peled · Answer 1 · 2019-02-25T14:48:23.613

3

If you want "real" downsampling, where each value is the mean of k values, you can use

unsampled_array.reshape(-1, k).mean(1)

Make sure unsampled_array is a np.array. In your case, k=2. That will give you:

[ 2. 6. 10. 14. 18.]

* Update: If you just want to take the first item for each k items, you can use this code:

unsampled_array.reshape(-1, 2)[:, 0]

Take a look at this plot:

edited Feb 25 '19 at 14:48

answered Feb 25 '19 at 02:33

Noam Peled

4,484
5
43
48

You are correct. I didn't mention it in the question but I was initially interested in downsampling objects and downsampling between and to sample sizes. I guess I over simplified the example. – Jesse Reza Khorasanee Feb 25 '19 at 02:53

score 2 · Accepted Answer · answered Feb 25 '19 at 02:31

2

You need function np.ix_, as follows:

import numpy as np


unsampled_array = np.array([1,3,5,7,9,11,13,15,17,19])
number_of_samples = 5
downsampling_indices = np.linspace(0, len(unsampled_array)-1, number_of_samples).round()
downsampling_indices = np.array(downsampling_indices, dtype=np.int64)

indices = np.ix_(downsampling_indices)
downsampled_array = unsampled_array[indices]

print(downsampled_array)

answered Feb 25 '19 at 02:31

Happy Boy

526
2
9

1

I don't think the `np.ix_` is necessary at all. You can just go `unsampled_array[ downsampling_indices]`. – nico Feb 25 '19 at 02:37
This is great as it also works with non integer down sampling ( ie 10->7 ) – Jesse Reza Khorasanee Feb 25 '19 at 02:43
1

`np.ix_` is very quick. You can compare both approaches. – Happy Boy Feb 25 '19 at 02:54
1

Instead of calling np.array with dtype=np.int64, add to the end of the np.linspace command astype(int). Also, instead of using round, I would use np.rint. – Noam Peled Feb 25 '19 at 03:08
1

@NoamPeled You are right about `astype(int)` but `np.rint` returns a float for some reason so there is no benefit of `np.rint(x).astype(int)` over `x.round().astype(int)` – nico Feb 25 '19 at 03:12
I prefer the "round half to even" approach when rounding floats to integers which are also used as indices. – Noam Peled Feb 25 '19 at 03:24
Instead of calculating the indices, just use this: unsampled_array.reshape(-1, 2)[:, 0] – Noam Peled Feb 25 '19 at 14:49
agree with you. i get it. – Happy Boy Feb 25 '19 at 14:52

How to down sample an array in python without a for loop

Minimum working example:

Result:

2 Answers2