The `argmin` method of the masked array is not working properly.

The argmin method of the masked array is not working properly.

import numpy as np

a = np.ma.array([1, 2, 1e+21, 1e+22], mask=[1, 1, 0, 0])
d = a.filled().view(np.ndarray)

print(d.argmin())  # 0
print(a.argmin())  # 2

However, the argmin method is implemented by computing d and then returning d.argmin():

# https://github.com/numpy/numpy/blob/master/numpy/ma/core.py
def argmin(self, axis=None, fill_value=None, out=None):
    if fill_value is None:
        fill_value = minimum_fill_value(self)
    d = self.filled(fill_value).view(ndarray)
    return d.argmin(axis, out=out)

So why I am getting two different results?

On the other hand, when the unmasked items are all inf’s, then the argmin returns always 0:

a = np.ma.array([1, 2, np.inf, np.inf], mask=[1, 1, 0, 0])
d = a.filled().view(np.ndarray)

print(d.argmin())  # 0
print(a.argmin())  # 0

But in both cases fill_value=1e+20 is smaller than a[2] = 1e+21 and a[2] = np.inf. In any case, the output I would expect is 2 for both examples, since I don’t want to compare the unmasked items with the masked items.

Numpy/Python version information:

1.15.3 3.6.4 |Anaconda, Inc.| (default, Jan 16 2018, 10:22:32) [MSC v.1900 64 bit (AMD64)]

[seberg]

Not sure, but maybe something like np.maximum(np.argmin(mask), np.argmin(filled)) might work. If the first unfilled value is actually larger then the found minimum, it must be an inf.

EDIT: The question is if that opens another bag of problems somehow…

[kechan]

I would love to do min and argmin on masked array as welll. I was looking for a solution in TF Tensor but landed here.