py03 how to use scipy csc matrix

[Liripo]

I would like to know how to use csc matrix in pyo3, for example, how to convert it into sprs CsMat?

[flying-sheep]

nothing yet I think: sparsemat/sprs#289

[ilgrad]

You can bridge the CSC storage arrays explicitly. SciPy and sprs both represent a CSC matrix using its shape, column pointers, row indices, and values.

First canonicalize a copy on the Python side:

A = A.tocsc(copy=True)
A.sum_duplicates()
A.sort_indices()

Then pass A.shape, A.indptr, A.indices, and A.data to your PyO3 function. For an owned real-valued matrix, convert them into (usize, usize), Vec<usize>, Vec<usize>, and Vec<f64>, respectively, and construct:

CsMat::<f64>::try_new_csc(shape, indptr, indices, data)

Use the fallible constructor and translate its error into a Python exception. SciPy index arrays may be int32 or int64, so handle both with checked conversion to the Rust index type. Likewise, explicitly support or reject the input value dtype; the example above assumes real f64 values.

This creates an owned copy. A borrowed CsMatView needs additional lifetime and mutation guarantees for the Python buffers.

References: SciPy CSC storage, sprs constructors.