Merge (Replace) one array to (by) other with dynamic size

[gsaha009]

Hello Experts,

I was trying to do an operation as described in the following:

I have two arrays

a = [ [0], [1], [1], [0] ]
b = [ [12.4], [23.5] ]

Now, I want to merge b to a where the value 1 matches. I mean I want to have something like this:
c = [ [], [12.4], [23.5], [] ]

The reason is the following:
I had some awkward arrays in the beginning. In order to do some existing operations, I had to convert it to numpy and, thats why I applied a mask to make that array regular. But, I saved the places where the new elements should be placed.

Please let me know if there is any way to do the same.

Thank you very much

Gourab

[agoose77]

If I understand your question, you have an array a that has been made regular (fixed-dimension). You have another array b that represents an "original" array whose contents need to be inserted back into a? Is that correct? i.e., is the number of 1 in your array a equal to the number of elements in array b?

It would be good to understand more about your problem; the shapes and dimensions of the arrays are important. But, we can make a start!
I only have a few minutes here, so I'm sure that @jpivarski or @pfackeldey will expand / provide alternatives.

What it seems like you want to do is to build an index that can pull out the proper elements of b with the right structure:

import awkward as ak
import numpy as np

flat_a = ak.flatten(a)
shape_a = ak.num(a)

flat_b = ak.flatten(b)

masked_flat_a = flat_a.mask[flat_a == 1]
flat_index = np.cumsum(masked_flat_a) - 1

flat_result = flat_b[flat_index]
result = ak.unflatten(flat_result, shape_a)
result

This code sample flattens both arrays (preserving the shape of a in shape_a), masks out the "valid" elements of a (so that the resulting array has None for the elements that are "padding" (0)), builds a zero-based index into b by counting each non-zero element of masked_flat_a, and then indexes into flat_b. Inspecting flat_index should show you how it works!

This solution uses Awkward's ragged indexing feature, which is very powerful.