Marc Glisse

> I am not sure about `pykeops`: outputing `num_landscapes` would require running that many "reductions", all applied to the same data. I think the API doesn't allow that. There is...

The CI tests don't pass. Could you have a look first? In particular the one called "build" that runs on CircleCI.

Hello, thanks for the report. I do not believe that we use the affected function, whether directly or indirectly, and I think packages are made in a way that the...

I don't know for sure that it is faster, one would need to benchmark it (and there is always a chance the 3d version might become faster again in the...

I just had a case where the number of simplices was 382, I ran collapse_edges, it remained 382, I reran collapse_edges, and it got down to 381. Just to mention...

> I think I would prefer a static method, ok. > Maybe harder to document. ? Static methods appear in the doc, with "static" in front. > I think we...

> Are you ok for `create_from_coo_matrix` to be static also ? It seems less obvious. While the other function requires to start from an empty simplex, this one does not,...

Thinking about it, SciPy's `coo_matrix` is a bit limiting in that it can only represent matrices (2d) and not arbitrary arrays. Making it more general, like `torch.sparse`, would allow inserting...

I am looking at this again, to see what really needs to be done before it can be merged (changing it would modify the API in a way that breaks...

I just added insert_batch, which does something similar to insert_edges_from_coo_matrix but for any dimension, and I optimized the code a bit more (the C++ code is not that cheap, so...