fix bug which happens when .mat data contains a sparse matrix in COO

[ruiying-ocean]

when fabmos (v 1.0.1 from conda-forge) uses a old version of TMM (MITgcm_ECCO), a bug occurs because the .mat data has COO (coordinate) format, but the Python code is expecting a CSR (compressed sparse row) format. This PR converts the format to avoid the bug.

Traceback (most recent call last):
  File "/Users/vhf24tbu/tmm/MITgcm_ECCO/PISCES.py", line 18, in <module>
    sim = fabmos.transport.tmm.Simulator(domain, calendar=calendar, fabm=fabm_yaml)
  File "/Users/vhf24tbu/.local/share/mamba/envs/fabmos/lib/python3.13/site-packages/fabmos/transport/tmm.py", line 480, in __init__
    self.Aexp_src = TransportMatrix(
                    ~~~~~~~~~~~~~~~^
        Aexp_files,
        ^^^^^^^^^^^
    ...<6 lines>...
        order=order,
        ^^^^^^^^^^^^
    )
    ^
  File "/Users/vhf24tbu/.local/share/mamba/envs/fabmos/lib/python3.13/site-packages/fabmos/transport/tmm.py", line 129, in __init__
    self._get_matrix_metadata(file_list[0], order=order)
    ~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/vhf24tbu/.local/share/mamba/envs/fabmos/lib/python3.13/site-packages/fabmos/transport/tmm.py", line 223, in _get_matrix_metadata
    A_ir = A.indices
           ^^^^^^^^^
AttributeError: 'coo_matrix' object has no attribute 'indices'

[jornbr]

Many thanks for reporting this and exploring a potential solution. It seems that recent scipy changes (scipy/scipy#21905) causes the matlab v5 arrays to be returned as coo_array, while previously they were returned as csc_matrix. That does mean, however, that the solution is to explicitly cast to CSC format, rather than the CSR format you propose here. In order to not depend further on details of scipy's sparse matrix implementation, I'd propose to add

A = scipy.sparse.csc_array(A)

just underneath

A = vardict[self._group_name]

That should address your issue and also maintain compatibilty with older scipy versions (in which case the cast does nothing). Could you test if that works for you, and if so, update your PR?

[jornbr]

Actually a cleaner code could likely just use

A = vardict[self._group_name].tocsc()

[ruiying-ocean]

Hi @jornbr, thanks for your rapid response!. your code is more efficient and it resolves my issue. I have updated the PR.

[jornbr]

CSR here should become CSC (or the comment could go, since the streamlined code is now pretty clear)

[ruiying-ocean]

absolutely! this is corrected.

[jornbr]

Perfect - thanks for your contribution!