A good sparse matrix implementation is key to any serious numerics. SciPy's implementation is very nice and quite complete, and a foundation for the entire scientific Python ecosystem. Unfortunately, though, many of the low-level sparse operations are not implemented very efficiently, which has lead to some projects having to do their own implementation, see e.g. this discussion for the QuTiP project:
<a href="https://github.com/qutip/qutip/issues/850#issuecomment-384005667" rel="nofollow">https://github.com/qutip/qutip/issues/850#issuecomment-38400...</a><p>It would be really nice if <i>all</i> of the sparse linear algebra in SciPy could be heavily optimized (to a similar extent as e.g. Intel optimizes their sparse operations in MKL), so that the entire ecosystem could benefit from that. This is probably something that would require some corporate support, but given how many data science and finance companies use Python for their workflow these days, it might be a wise investment.
There seems to be no end of sparse formats, potential sparse formats and algorithms for dealing with them.<p>But it also seems like for a given sparse matrix in a given situation, there's no guaranteed that there's algorithm for handling it. The whole thing requires deep experience or black magic.