Replace division by zero numpy

Question

I'm doing a matrix by matrix pointwise division however there are some zeros in the divisor matrix. This results in a warning and in some NaNs. I want these to map to 0, which I can do like this:

edge_map = (xy/(x_norm*y_norm))
edge_map[np.isnan(edge_map)] = 0

However there are two issues with this, first of all it still gives a warning (I don't like warnings) and second of all this requires a second pass over the matrix (not sure if this is unavoidable) and efficiency is very important for this part of the code. Ideas?

Answer 1

This is probably the fastest solution, but the where function does trigger the error as it precalculates the solutions:

import numpy as np

n = 4

xy = np.random.randint(4, size=(n,n)).astype(float)
x_norm = np.random.randint(4, size=(n,n)).astype(float)
y_norm = np.random.randint(4, size=(n,n)).astype(float)

xy_norm = x_norm*y_norm

edge_map = np.where(xy_norm == 0, xy_norm, xy/xy_norm)

print(xy)
print(xy_norm)
print(edge_map)