Erik Welch

I agree with your points @szhorvat, and thanks for the clear explanations. It's helpful to have a "paper trail", and I'm looking forward on your perspective on more things. We'll...

Perhaps the easiest thing for us to do here (at least for now) is to double the diagonals ourselves when converting from NetworkX, such as: ```python if not G.is_directed(): A...

If we double the diagonals by default when converting _from_ NetworkX, then we ought to halve the diagonals when converting _to_ a NetworkX undirected Graph by default. What are good...

Thanks for the suggestions, questions, and heads up @szhorvat. I expect to be especially busy the next two weeks, so this is good timing for a pause. I do appreciate...

@jim22k, regarding the shape of the incidence matrix, since we assume row-oriented-ness, I think it makes the most sense to have `nodes X edges` (node ids as row indices and...

@SultanOrazbayev @jim22k for the next meeting or two, let's try to work out a proposal for going to/from networkx and adjacency matrix. If we have the time, I think it...

The relationship I'm more familiar with the relationship connecting the incidence matrix `B` to the Laplacian `L`, adjacency matrix `A`, and diagonal degree matrix `D`. For unweighted, undirected graph: ```...

Heh, we'll find out. I should have mentioned `dimsum` that @jim22k experimented with a few years ago to provide a N-dimensional wrapper on top of GraphBLAS: https://github.com/jim22k/dimsum SuiteSparse:GraphBLAS provides some...

Some places this has come up: - https://github.com/Transurgeon/cvxpy/pull/42 - https://github.com/Nixtla/neuralforecast - GraphBLAS backend implementation of a model in https://github.com/huggingface/transformers

> However, it does feel like something that should be a wrapper/companion package, rather than python-graphblas feature. That's a good point. It probably makes more sense to create a new...