TypeError for RegularGridInterpolator with big-endian data

[cdeil]

I ran into this TypeError with the RegularGridInterpolator for some big-endian data:

>>> import numpy as np
>>> from scipy import interpolate
>>> 
>>> values = np.ones((10, 20 , 30), dtype='>f4')
>>> points = map(np.arange, values.shape)
>>> xi = [(1, 1, 1)]
>>> interpolator = interpolate.RegularGridInterpolator(points, values)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/scipy/interpolate/interpolate.py", line 1495, in __init__
    raise ValueError("fill_value must be either 'None' or "
ValueError: fill_value must be either 'None' or of a type compatible with values
>>> interpolator = interpolate.RegularGridInterpolator(points, values, fill_value=0.)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/local/Library/Frameworks/Python.framework/Versions/2.7/lib/python2.7/site-packages/scipy/interpolate/interpolate.py", line 1495, in __init__
    raise ValueError("fill_value must be either 'None' or "
ValueError: fill_value must be either 'None' or of a type compatible with values

I think this is a small bug because the Python float 0. should be compatible with a dtype of >f4, no?

cc @andreas-h @pv

[cdeil]

Also it took me a few minutes to figure out what to pass for points ... I guess using "pixel coordinates" is a very common use case and this could be shown in an example?

points = map(np.arange, values.shape)

Am I correct that there is no 3-dim spline interpolation available in scipy?

[andreas-h]

thanks for reporting, christoph!

@pv I'm by no means a dtype expert ... can you give me a hint how to tackle this?

[pv]

Maybe can_cast(..., casting='same_kind'). This may be need to be
guarded with Numpy version checks, as the kwarg is added only in Numpy
1.6 (i.e. don't make any checks at all for Numpy < 1.6). The intent is
just to catch early errors due to assigning floats to integers or
complexes to floats.

Do version check by adding a global variable NUMPY_LT_160 = NumpyVersion(np.__version__) < '1.6.0' (needs to be global as
NumpyVersion is a slow operation).