Up until now I hadn't actually used my new-ish mask keyword argument feature. All the tests passed so I assumed it would work. I've now started to use it in my own work and discovered that it doesn't actually work for most real world cases because it has a major bug.
What was actually happening is that my code was getting lucky. My code was checking the query's mask array at index start_idx + i but it should have been using pidx[start_idx + i] which is the actual index of the source data (and the mask array). The previous index I was using was actually the index in the KDTree structure (I think). In all of my tests I was getting lucky because the KDTree happened to place the data points in the same positions in its structure as the original array so both start_idx + i and pidx[start_idx + i] were the same.
I've fixed the code and updated my "fake data" tests so that the input data are in a random order instead of sorted order. The test1d_mask test fails before my changes and passes after. It looks like the real world data in data_pts_real are sorted enough that test3d_mask passes with/without my fix.
Up until now I hadn't actually used my new-ish
maskkeyword argument feature. All the tests passed so I assumed it would work. I've now started to use it in my own work and discovered that it doesn't actually work for most real world cases because it has a major bug.What was actually happening is that my code was getting lucky. My code was checking the query's
maskarray at indexstart_idx + ibut it should have been usingpidx[start_idx + i]which is the actual index of the source data (and the mask array). The previous index I was using was actually the index in the KDTree structure (I think). In all of my tests I was getting lucky because the KDTree happened to place the data points in the same positions in its structure as the original array so bothstart_idx + iandpidx[start_idx + i]were the same.I've fixed the code and updated my "fake data" tests so that the input data are in a random order instead of sorted order. The
test1d_masktest fails before my changes and passes after. It looks like the real world data indata_pts_realare sorted enough thattest3d_maskpasses with/without my fix.