5

Is there a way to "expand" an array and repeat the last element to fill the expansion?

Another post talks about expansion and padding with 0 but I wish to repeat the last value as the pad.

Say I have an array:

[[1, 2],
 [3, 4],
 [0, 0]]

And I need to insert [5, 6, 6] to replace the [0, 0], obviously NumPy wouldnt allow this. But can I reshape/expand to:

[[1, 2, 2],
 [3, 4, 4],
 [5, 6, 6]]

I'm reading through a file where the number of values may vary in length, but I need the array to be of the same shape. One way to do this is read through the file first and find the maximum length, then read it again an populate, but the file is 10GB+ so I would prefer to do it on a single pass by "expanding" and backfilling with repeats.

pstatix
  • 3,611
  • 4
  • 18
  • 40
  • From the link you shared: https://stackoverflow.com/a/25173755/11301900 – AMC Feb 26 '20 at 18:12
  • @AMC Wasn't aware that `np.pad` offered an edge mode. – pstatix Feb 26 '20 at 18:21
  • Does this answer your question? [Good ways to "expand" a numpy ndarray?](https://stackoverflow.com/questions/12668027/good-ways-to-expand-a-numpy-ndarray) – AMC Feb 27 '20 at 17:29

1 Answers1

4

Looks like what you require is numpy.pad using the edge mode. From the doc:

‘edge’
    Pads with the edge values of array.

Example code:

>>> ar = np.array([[1,2], [4,5]])
>>> ar
array([[1, 2],
       [4, 5]])
>>> np.pad(ar, [(0, 0), (0, 4)], mode="edge")
array([[1, 2, 2, 2, 2, 2],
       [4, 5, 5, 5, 5, 5]])

The first (0, 0) tuple specify no padding on the first axis, while the second basically does "add 0 padding to the left and 4 to the right"

Cal
  • 631
  • 3
  • 11