Considering removing gensim as a dependency

[bdpedigo]

Considering completely removing gensim (and therefore, node2vec) as a dependency in graspologic

They constantly have dependency issues because the package is not heavily maintained (currently releasing on PyPI about once per year). For instance, right now builds are failing due to an issue with Scipy. piskvorky/gensim#3525

If anyone really wants to keep node2vec around, then I'd like to discuss how to get away from these dependency conflict issues or to have someone help with maintenance when they arise

[bdpedigo]

cc @daxpryce @darthtrevino: I know you're likely busy with other things, but if you two know of anyone else at MSFT using graspologic that might care, could you please cc them as well? Thank you in advance

[daxpryce]

I believe that @darthtrevino and company still use it, but I would agree it has become such a millstone around the neck re: library versions, python version limitations, etc, that you're better off removing it until a... less challenging version can be built or found and used. What do you think @darthtrevino ?

[darthtrevino]

We mostly use this library for Leiden in GraphRAG, but we do have node2vec as an optional feature. We've been discussing moving to GEE. Is there a way to support n2v without gensim? Maybe an alternative implementation?

[bdpedigo]

this one popped up for me with 1.2k stars https://github.com/eliorc/node2vec I have also definitely seen other libraries

[darthtrevino]

If possible, we'd prefer to have the n2v embedding operation supported, but we're not married to gensim per se

[daxpryce]

I would assert the main reason we added node2vec in topologic and then graspologic is specifically so we could swap out implementations but provide a consistent API. I think the effort to replace it rather than remove it is good, though I also could see removing it in the next couple versions until you can get the next one in place - if only so development can continue without dependency requirement conflicts

[bdpedigo]

looks like things are working now https://github.com/graspologic-org/graspologic/actions/runs/11259180298

i somehow missed that they had published a fix on PyPI for the current issue.

going to leave this issue open though, i think this is a general problem that we have a pretty wide scope of dependencies