Inconsistent behaviour nd.argmax against np.argmax when there are 'nans' in data

[access2rohit]

below is a reproducable example to display inconsistencies in argmax in mxnet and numpy

>>> import mxnet as mx
>>> import mxnet.ndarray as nd
>>> x = np.array([[1,5,3],[float('nan'), 2,6]])
>>> x
array([[ 1.,  5.,  3.],
      [nan,  2.,  6.]])
>>> y = nd.array(x)
shape=(2, 3)
[14:05:53] src/c_api/c_api.cc:190: MXNDArrayCreateExInt64 shape=2
>>> nd.argmax(y, axis=0)
[0. 0. 1.]
<NDArray 3 @cpu(0)>
>>> nd.argmax(y, axis=1)
[1. 2.]
<NDArray 2 @cpu(0)>
>>> import numpy as np
>>> np.argmax(x, axis=0)
array([1, 0, 1])
>>> np.argmax(x, axis=1)
array([1, 0])

[mxnet-label-bot]

Hey, this is the MXNet Label Bot.
Thank you for submitting the issue! I will try and suggest some labels so that the appropriate MXNet community members can help resolve it.
Here are my recommended label(s): Bug

[access2rohit]

@mxnet-label-bot add [numpy]

[access2rohit]

@reminisce @szha

[haojin2]

@access2rohit Seems like this issue has been addressed in the latest nightly build (mxnet-mkl==1.6.0b20191229):

>>> import mxnet as mx
>>> import mxnet.ndarray as nd
>>> import numpy as np
>>> x = np.array([[1,5,3],[float('nan'), 2,6]])
>>> y = nd.array(x)
>>> nd.argmax(y, axis=0)

[1. 0. 1.]
<NDArray 3 @cpu(0)>
>>> nd.argmax(y, axis=1)

[1. 0.]
<NDArray 2 @cpu(0)>

Closing this issue now, if you found the problem is not solved please feel free to re-open this issue.