Erik Welch

> Dijkstra's algorithm requires all edges to have positive weights. In order to support weight 0 edges and negative weight edges we need to implement the Bellman-Ford algorithm. The GraphBLAS...

This looks like a reasonable approach. The latest dev version of `np.linalg.inv` [says this](https://numpy.org/devdocs/reference/generated/numpy.linalg.inv.html): > If a is detected to be singular, a [LinAlgError](https://numpy.org/devdocs/reference/generated/numpy.linalg.LinAlgError.html#numpy.linalg.LinAlgError) is raised. If a is ill-conditioned,...

Thanks for the question. Is this a duplicate of #550? I left a reply there. Sorry for the delay around the holidays; #550 must have slipped under the radar.

Sorry you bumped into this! We are still working to get numpy 2 support in #546, which I think is getting close. The error you are seeing originates from `suitesparse_graphblas`,...

How interesting, thanks for sharing! I don't believe SuiteSparse:GraphBLAS currently uses MKL or any dense acceleration libraries for dense or bitmap operations, but perhaps someday it could. CC @DrTimothyAldenDavis, who...

I updated conftest.py so that the `nx-loopback` backend is only present when testing a backend, so normal testing should now go through the fast path. This change required `nx-loopback` to...

The dispatch CI test is failing because `apply_matplotlib_colors` creates tuples of numpy scalars, which gets checked (and raised) downstream by `get_edge_attributes`. **edit:** to fix this, update so that functions don't...

Okay, I fixed it by changing `apply_matplotlib_colors` to not create tuples of numpy scalars.

Thanks for the friendly ping (and for working on this PR) @amcandio! I am eager to help get this in. However, I have very limited bandwidth this summer, so I...

FYI: somewhat related, we have, and are, exploring caching results of functions (see #7283). For example, this can allow us to compute connectedness of a graph only once, which may...