[Numpy] The argument parser of some operators cannot parse numpy integers or mx.numpy integers correctly.

[sxjscience]

import mxnet as mx
import numpy as np
mx.npx.set_np()
a = mx.np.ones((10, 10))
b = mx.np.split(a, np.int32(1))

Error:

ValueError: indices_or_sections must either int, or tuple / list / set of ints

The reason is that we are testing against the python int here:
https://github.com/apache/incubator-mxnet/blob/4f14bf4fe51b586f86a1e7bd38b8282997e6ffbb/python/mxnet/ndarray/numpy/_op.py#L2733

One way to solve the problem is to see if the indices_or_sections belongs to integer_types:
https://github.com/apache/incubator-mxnet/blob/4f14bf4fe51b586f86a1e7bd38b8282997e6ffbb/python/mxnet/base.py#L49

However, this will still be problematic, because the interger_types do not include the mxnet.numpy integers, i.e the following script will still raise the error:

import mxnet as mx
import numpy as np
mx.npx.set_np()
a = mx.np.ones((10, 10))
b = mx.np.split(a, mx.np.int32(1))

We will need some one to look at the problem @reminisce @haojin2 .

[sxjscience]

I'll partially solve the problem for the split/hsplit/vsplit operators by using integer_types in #16621. However, we need a way to support mxnet numpy scalars.

[sxjscience]

After discussion with @junrushao1994, we can try to convert the input to python int first. Also, in JAX and numpy, it supports float32 inputs and automatically detects the float case.

import jax.numpy as np
a = np.ones((10, 10))
b = np.split(a, np.float32(2))

import jax.numpy as np
a = np.ones((10, 10))
b = np.split(a, np.float32(2.2))
# Raises ValueError: array split does not result in an equal division