This PR eliminates a for loop in preference for a single vectorized numpy call. This saves 20-30 s from each fiberassign run. In combination with skipping QA steps and replacing yaml with json in load_focalplane, this reduces the time to get fiberassign files from something like 90 s to something like 20 s.
I still need to verify that this produces identical output. I'll try to get to this tomorrow. But it would very much surprise me if it did not.
This PR eliminates a for loop in preference for a single vectorized numpy call. This saves 20-30 s from each fiberassign run. In combination with skipping QA steps and replacing yaml with json in load_focalplane, this reduces the time to get fiberassign files from something like 90 s to something like 20 s.
I still need to verify that this produces identical output. I'll try to get to this tomorrow. But it would very much surprise me if it did not.