numpy.maximum() on two 3-D arrays

Question

I'm looking for the numpy-way to compare two 3-D arrays and return a new 3-D array containing the element-wise maxima based on a given index of the array. Here my value is in [y][x][7] for instance:

output = numpy.zeros((16, 16, 8))
# we want to compare the values in [y,x,7] and only keep the element with max value
for x in range(array_min.shape[1]):
  for y in range(array_min.shape[0]):                
    if abs(array_min[y][x][7]) > array_max[y][x][7]:
      output[y][x] = array_min[y][x]
    else:
      output[y][x] = array_max[y][x]

score 2 · Accepted Answer · answered Jun 11 '20 at 00:15

2

If I understand you correctly, you only want a specific index on 3rd dimension to be compared. In that case, numpy has a builtin function for this that you can replace your loop with:

output[:,:,7] = np.maximum(array_min[:,:,7], array_max[:,:,7])

answered Jun 11 '20 at 00:15

Ehsan

12,072
2
20
33

Thanks, would there be a way to add the absolute value in the max(abs(min), max)? – Roms Jun 11 '20 at 15:56
Also, I want to keep the same dimension in the output array. For instance here output.shape should still be (16,16,8). – Roms Jun 11 '20 at 16:30

tstanisl · Answer 2 · 2020-06-12T12:25:52.143

1

output = np.where((abs(array_min[...,7]) > array_max[..., 7])[..., None], array_min, array_max)

edited Jun 12 '20 at 12:25

answered Jun 10 '20 at 22:22

tstanisl

13,520
2
25
40

1

This leads to "operands could not be broadcast together with shapes (16,16) (16,16,8) (16,16,8) ". Am I missing something? – Roms Jun 11 '20 at 16:32

Grigory Feldman · Answer 3 · 2020-06-10T22:31:26.673

1

array_min = np.random.rand(16,16,8)
array_max = np.random.rand(16,16,8)
out = np.stack([array_min, array_max], axis=0).max(axis=0)

Works for more than 2 arrays.

edited Jun 10 '20 at 22:31

answered Jun 10 '20 at 22:25

Grigory Feldman

405
3
7

numpy.maximum() on two 3-D arrays

3 Answers3