Data comparison is inconsistent with numpy

[marongxing]

In [88]: nd.array([0],dtype='int') < 1e-3
Out[88]:

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

In [89]: np.array([0],dtype='int') < 1e-3
Out[89]: array([ True])

In [90]:

[haojin2]

Hi,
The behavior of nd is not guaranteed to be consistent as official NumPy.
If you're looking for NumPy-compatible behaviors, please try out the new NumPy-compatible interface, available in latest nightly versions & soon to be available in 1.6.0 release version:

>>> from mxnet import np
>>> np.array([0],dtype='int') < 1e-3
array([False])

[ptrendx]

@haojin2 But the problem is not the datatype, but the result - 0 is less than 1e-3 ;-).

[haojin2]

@ptrendx I think it's a data type issue:

>>> from mxnet import np
>>> np.array([0],dtype='float') < 1e-3
array([ True])

I guess 1e-3 was casted to integer in the backend...

[sxjscience]

I think the root cause would be the same as #16653

[yzhliu]

@Tommliu is working on this.

[yzhliu]

closed by #18277