For ease of explanation I will be using 2-dimensional numpy arrays (I am using 3-dimensional arrays in my problem).

I have 2 arrays, one records the odds of something occuring at those specific coordinates. The other array is a pre-generated matrix that is used to lower the values of the first array around a central point in a pre-determined radius.

I want to automatically select points in the first matrix (henceforth A), and prevent the program to select another point that sits too close to previously selected points. So I want to multiply the values around the selected point related to the distance from said point.

E.G.:

Matrix A:

[[0, 0, 0, 0, 0, 0], [0, 1, 2, 2, 1, 0], [0, 2, 4, 4, 2, 0], [0, 2, 4, 4, 2, 0], [0, 1, 2, 2, 1, 0], [0, 0, 0, 0, 0, 0]]

Matrix B:

[[ 1, 0.5, 1 ], [0.5, 0, 0.5], [ 1, 0.5, 1 ]]

Now say that index [2, 1] is selected as a point of interest. B is multiplied with A, but only with the values in a 3*3 around [2, 1]

Result:

[[0, 0, 0, 0, 0, 0], [0, 0.5, 0, 1, 1, 0], [0, 2, 2, 4, 2, 0], [0, 2, 4, 4, 2, 0], [0, 1, 2, 2, 1, 0], [0, 0, 0, 0, 0, 0]]

This should result in the points around [2, 1] not being valuable enough to be selected as points of interest, unless the conditions are high enough to be selected anyway, hence the multiplication.

Now, I can’t seem to figure out a way to perform this specific multiplication. `numpy.multiply()`

would repeat B so that it gets applied over the entire matrix A while I only want to apply it on a small part of A.

Another option would be looping over the affected section of Matrix A, but this requires an insane amount of time (especially in 3-dimensional matrices)

In other words, I want to apply a convolution filter without summing up the multiplication results at the end, but apply them to the underlying values of the convolved matrix (A)

Any insights on this issue are appreciated.

## Answer

The easiest solution uses slicing:

A[5:8, 8:11] = np.multiply(A[5:8,8:11], B)

What this does is extract from A the 3×3 area around the selected point (here [6,9]), multiply it (element-wise) by B, and write it back at the same location.

Since you talked about convolution, if you want to use that the approach would be to create a matrix M of the same shape as A, but that is zero everywhere except at the selected point. This matrix you can convolve with B, and then multiply with A:

M = np.zeros(A.shape) M[6,9] = 1 M = scipy.ndimage.filters.convolve(M, B) A = np.multiply(A, M)

(or something to that extent, I did not test this variant).