inconsistent return value of find_k_nearest_gpi

[sebhahn]

Depending on whether the NN search returns a distance containing a np.inf or not decides whether the returned indices and distances are a 1d array (mask applied) or a 2d array (no mask applied, i.e. no np.inf entry in the distances)

https://github.com/TUW-GEO/pygeogrids/blob/master/src/pygeogrids/grids.py#L468

[wpreimes]

should we make it consistent?

[sebhahn]

yes I think so, however, I'm not sure if a 2d array is the actual intention

according to the tests it might be, I'm just going through them

[wpreimes]

I think I changed that part of the code some time ago, and I made sure that it works the same way as before. But I also was not sure at that time if it was on purpose or not. I think it can always be 1d.

[sebhahn]

no, I think it has to be 2d because it is possible to pass multiple lat/lon coordinates and get k neighbors in return.
using a boolean mask reduces the 2d array to a 1d array, so one solution could be keeping also distances with np.inf otherwise the number of neighbors might be different querying two coordinates at the same time

[wpreimes]

ah ok, makes sense

[sebhahn]

I hope changing this behavior is not breaking any code

[sebhahn]

@wpreimes how should we handle np.inf in case of find_nearest_gpi?

a) return np.inf
b) or return empty array/None?

[wpreimes]

fixed in #76