New keyword parameter to cf.Field.regrids and cf.Field.regridc: max_masked

[davidhassell]

Fixes #949

All the excitement happens in cf/data/dask_regrid.py. The rest is just passing the new max_masked parameter about, docs and tests :)

[sadielbartholomew]

Aside from some typos it is perfect, so please review those and then merge at will.

That said, I wanted to comment regarding API consistency - though it relates more to the issue at hand and your proposed way to handle it (i.e. #949) rather than the implementation of it here. Namely I notice we have the keyword 'mtol' for Field.collapse and the various convenience methods for collapses (mean, integral etc.) and that aligns very closely with the new max_masked keyword introduced, given its description:

Set the fraction of input data elements which is allowed to contain missing data when contributing to an individual output data element. Where this fraction exceeds mtol, missing data is returned. The default is 1, meaning that a missing datum in the output array occurs when its contributing input array elements are all missing data. A value of 0 means that a missing datum in the output array occurs whenever any of its contributing input array elements are missing data. Any intermediate value is permitted.

They have different names but (by my understanding) effectively control the same thing (input masking to output masking) - bar the slight complication of the role of any weighting used for the regrid case? But one is specified by number, the other by fraction.

So in future we could consider choosing one over the other for both collapses and regridding (and any other applicable methods).

[davidhassell]

Hi Sadie - thanks for the review.

I'd like to think about the pertinent API question you raise before merging. mtol specified as a fraction in [0,1] has the advantage that for linear regridding from unstructured source grids, there may be different numbers of potentially contributing cells for each destination grid cell. So being able to specify a fraction (0.75, say) means you can play nicely with triangles, squares, pentagons, etc.

This would involve small tweak to the code (to ascertain the number of source grid cells in play, and to convert the fraction to a number of masked points in each case), but that's no more or less fiddly than what we've already got!

I'll take a look ....

[davidhassell]

Hi Sadie - PR updated for mtol ... 698b18d

[sadielbartholomew]

Thanks David for updating the approach - sorry it was extra work but like you indicated, it provides a more flexible approach than the max_masked original kwarg idea and it means we have a consistent kwarg 'mtol' across multiple supported methods. So, much nicer overall.

I reviewed the new commit and then did a sanity check on the whole PR branch. Overall is all good. Some more typos to correct (just two really, copied and pasted across) and a changelog conflict to resolve, but other than that all ready to merge so please go ahead.

[davidhassell]

Thanks, Sadie!