Repository navigation
ellipses (...) as indices on numpy array: #827
Description
Activity
Wrapped Julia arrays intentionally only support some very basic behaviour. If you want more array functionality you should convert it to a specialised array type like a numpy array.
Your
test_ellipsis(X)function implicitly assumes thatXis a numpy array, so you should pass one, like your suggested work-around.Alternatively the function could be modified to take more general array-like types by converting
Xitself to a numpy array, but you said that you cannot change the python code.Your specific suggestion of supporting
...wouldn't even help because it would then error on.sum()which also isn't implemented for wrapped Julia arrays.Reacted by Alexander BarthOk, Thank you for clarifying that. Actually, in my case, all python functions that I try to wrap essentially assume that arguments are numpy arrays. Is there a way to make the conversation automatic? Or would it be necessary to wrap the arguments manually?
PythonCall doesn't have any machinery for this directly.
Easiest is prob to make a little arg-making function like
makearg(x::AbstractArray) = pyimport("numpy").asarray(x) makearg(x) = Py(x) callfunc(f, args...) = f(map(makearg, args)...)
then call like
callfunc(script.test_ellipsis, randn(2, 2, 2))
or something along these lines.
Thanks a lot for your help!
I am trying to wrap a python code from a colleague. I would like to avoid any changes to the python code that makes the python code less general. I have a problem when the code uses ellipses (
...) as indices on numpy array:Is your feature request related to a problem? Please describe.
Consider the python code (in file
test_ellipsis.py) :The code is called in julia as:
Unfortunately, the produces the error:
Describe the solution you'd like
Could indexing of ellipses be supported in PythonCall.jl for numpy arrays ?
Describe alternatives you've considered
I could create a np.array from the julia side, for example:
Additional context
I am using julia 1.12 on Linux with PythonCall v0.9.31 and Numpy 2.5.3.
Btw, thanks for the great package!!