Closed ctuguinay closed 1 month ago
Closing this because it is a non-issue. I was thinking about this over the weekend, and regardless of which one is faster, we should develop upon the option that is the 'most sound'. There are 3 problems I already see for the second option:
apply_ufunc
with exclude_dimensions
, meaning broadcasting and alignment cannot be used here, which is one of the ways that apply_ufunc
makes operations fast (and potentially take up less RAM).
Related to the merging of #1314, the upsampling operation, which assumes non-uniform depth, for impulse noise has two possible implementations that were put forth in the revisions of the PR:
Assignment via Reindex:
This option is what is currently in
echopype
.Assignment via
.apply_ufunc
using upsample mapping:This issue is to determine which of these is faster. Potentially, the assignment via reindex could also be done using
apply_ufunc
using Pandasreindex_like
, and if this can be used in the 1st option, then the 1st option would most likely beat the 2nd option.