numpy.float() returns non-numpy python builtin float

numpy.float() returns a non-numpy python builtin float type (same goes for numpy.int()):

>>> print type(np.float(0))
<type 'float'>
>>> print type(np.int(0))
<type 'int'>

This is different to behavior of dtype options in array initialization:

>>> print type(np.arange(2, dtype=np.float)[0])
<type 'numpy.float64'>
>>> print type(np.arange(2, dtype=np.int)[0])
<type 'numpy.int64'>

I find this unexpected and surprising. This may have little practical impact as it only affects initialization of single numbers, but it means that e.g. numpy floating point operation error handling is circumvented in this case (which is how I stumbled upon this):

>>> np.seterr(all="ignore")
{'over': 'ignore', 'divide': 'ignore', 'invalid': 'ignore', 'under': 'ignore'}
>>> np.geterr()
{'over': 'ignore', 'divide': 'ignore', 'invalid': 'ignore', 'under': 'ignore'}
>>> np.float(0) / np.float(0)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ZeroDivisionError: float division by zero

[eric-wieser]

#6103 suggests deprecating these, for those following the trail

[rkern]

Specifically, numpy.float and numpy.int are only present because of backwards compatibility with a very early version of numpy that exposed numpy.float64 and numpy.int32 (or np.int64 on appropriate 64-bit platforms) under those names. It was soon realized that from numpy import * caused the builtin float and int to be obscured. However, because some people had started using dtype=numpy.float explicitly, we kept the name aliased to float such that it would continue working.

Please do not use these aliases. If you see documentation using them, please let us know so we can correct it.