nanarg* broken

[dcherian]

from numpy_groupies.aggregate_numpy import aggregate

array = np.array([1, 1, 2, 1, 1, np.nan, 6, 1])
labels = np.array([0, 0, 0, 0, 0, 0, 0, 0])

func = "nanargmax"
actual = aggregate(labels, array, func=func).astype(int)
expected = getattr(np, func)(array)
np.testing.assert_equal(actual, expected)

fails with

AssertionError: 
Arrays are not equal

Mismatched elements: 1 / 1 (100%)
Max absolute difference: 1
Max relative difference: 0.16666667
 x: array([5])
 y: array(6)

I think the issue is that you can't use nansqueeze with nanarg* reductions:

numpy-groupies/numpy_groupies/aggregate_numpy.py

Lines 272 to 275 in 7a31d1f

	if _nansqueeze:
	good = ~np.isnan(a)
	a = a[good]
	group_idx = group_idx[good]

Or you could create a vector of indexes, apply the nan mask, and then index that vector with the current nanarg* output.

[ml31415]

Hi Deepak! Thanks for your valuable input. So far I can confirm all your findings, and I'm a bit ashamed, that our test suite obviously has some gaps. PRs on that stuff would be highly welcome, not sure if I get to fix those errors in a timely manner.

[ml31415]

@dcherian please have a look if these fixes were sufficient for you.