Compute weighted sum of a 2D Numpy array

Question

I have a numpy 2D array of some dimension say 2 by 2 (in numpy.float32 dtype)

[[0.001 0.02],
 [0.3 0.9]]

I want to create a new mask matrix for the given matrix such that,

if a1<= matrix element <a2:
    new element = a3
if b1<= matrix element <b2:
    new element = b3
... Can have up to 10 conditions like this

For example if the conditions are:

if 0.0<= matrix element <0.05:
    new element = 10
if 0.05<= matrix element <1.0:
    new element = 5

The given matrix should transforms to:

[[10.0 10.0],
 [5.0 5.0]]

Can someone please help me to obtain the mask matrix for any given matrix with given set of conditions?

Usecase Basically instead of just using np.sum(numpy 2D array), I want to do a weighted sum of matrix elements. So the conditions are actually defining the weights.

Once having the original matrix and the mask matrix I shall then do element-wise multiplication of both matrices and then use np.sum on the resulting 2D array.

Probably would like to have less for loops as possible for fast execution. I am using Python 3.7.

This post is similar but do not really understand if that solves my task.

Thankyou

score 3 · Accepted Answer · edited Jul 06 '20 at 15:17

3

You can achieve this result with numpy.select and a dictionary summarizing your conditions.

arr = np.array([[0.001, 0.02], [0.3, 0.9]])
selection_dict = {10: (0<=arr)&(arr<0.05),
                  5:  (0.05<=arr)&(arr<1.0)}

In [52]: np.select(condlist=selection_dict.values(), choicelist=selection_dict.keys())
Out[52]: 
array([[10, 10],
       [ 5,  5]])

edited Jul 06 '20 at 15:17

marc_s

732,580
175
1,330
1,459

answered Jun 10 '20 at 13:05

AMH

502
2
10

That dictionary is a bit redundant, no? :-) – Nils Werner Jun 10 '20 at 13:10
I am not sure what you mean by "redundant" – AMH Jun 10 '20 at 13:11
1

You're creating a dict, just to take it apart again in the next line. – Nils Werner Jun 10 '20 at 13:11
1

Oh right ! I prefer to use a dict rather than two lists, because it keeps the conditions on the same line as the values. With two lists, if you have something like 20 conditions you might shift one by mistake – AMH Jun 10 '20 at 13:13
1

Could you please accept the answer if this helped you, thanks ! – AMH Jun 10 '20 at 14:13

Compute weighted sum of a 2D Numpy array

1 Answers1