✨ Feature Request

Enable the fast_percentile_method of the PERCENTILE aggregator to work on masked arrays if mdtol is set to zero.

Motivation

fast_percentile_method is, well, fast. So it would be good to be able to use it in a wider range of contexts. The particular use-case I'm thinking of is where all land points (and no other points) are masked and we aggregate over all the dimensions except the spatial ones. So for the grid cells we are interested in, we have a full (non-masked) set.

Additional context

The fast_percentile_method of the PERCENTILE aggregator uses numpy.percentile, which just ignores any mask, so in general we'd get the wrong values out:

import numpy as np
import numpy.ma as ma

arr = ma.array(np.tile(np.arange(6), 2).reshape(2, 6))
arr[0, 4:] = ma.masked

print(arr)
print(np.percentile(arr, 50, axis=1))

[[0 1 2 3 -- --]
 [0 1 2 3 4 5]]
/net/project/ukmo/scitools/opt_scitools/conda/deployments/default-2022_04_05/lib/python3.8/site-packages/numpy/lib/function_base.py:4650: UserWarning: Warning: 'partition' will ignore the 'mask' of the MaskedArray.
  arr.partition(
[2.5 2.5]

So we currently just don't allow it to run on masked arrays:

However, if mdtol was set to zero, we would always add a mask over those wrong values in the result:

or

So I think it would be safe to relax the check so that this method can still run if mdtol is zero.