LFC fails with NaN

[jrleeman]

When sending lfc data with NaNs (i.e. IA State), we get None as the result

[dopplershift]

But...masked arrays?

[jrleeman]

Yep - it is fine with a toy example, but has issues with real data. Trying to isolate the issue.

[jrleeman]

Example... what the? https://gist.github.com/jrleeman/f2e43c81d26741de39b175c280cb2ba5

[jrleeman]

Note to self/all - turns out it only fails if the masked point brackets the intersection. Trying to come up with a better implementation now.

[jrleeman]

This needs cleaned up, but I'm thinking of a utility like this - still may have problems depending on where the masked points are. More testing and thinking to do there.

def next_non_masked_element(a, idx):
    if hasattr(a, 'mask'):
        next_idx = idx
        while ma.is_masked(a[next_idx]):
            if next_idx < (len(a)-1):
                next_idx += 1
            else:
                break
        if ma.is_masked(a[next_idx]):
            return None, None
        else:
            return next_idx, a[next_idx]

I don't like simply or'ing masks together and we will loose real data that can improve the estimate.

[jrleeman]

A direct trickle down of #351 - closing and following the PR related to that issue.