Incompatibility with some version(s) of scipy

[balbasty]

It has come up during the SPM python course that the sparse arrays we use (scipy.sparray) were only introduced in scipy 1.11. This leads to errors when scipy < 1.11 is used.

[johmedr]

Should we just update the import statement for scipy.sparse?

try:
    from scipy import sparse
    from scipy.sparse import sparray # try to get sparray here? 
except (ImportError, ModuleNotFoundError):
    sparse = None

[balbasty]

That sounds good. Maybe also check for sparse.csc_array and sparse.coo_array, as they are also used in the code?

Although in practice it seems that sparray is only used here and here. sparray is the sparse base class, which existed already in < 1.10 but was private (_sparray). So if we want to cover more versions of scipy, we could not use it as a base class to WrappedSparseArray when it is not available, and use the more inconvenient test

if not isinstance(arg, (np.ndarray, sparse.coo_array, sparse.csc_array, ...))

Does Matlab support more than 2 dimensions for sparse arrays? If not, we could also replace sparray with spmatrix, which was already public in scipy 1.10