SMMR

[do-me]

Great repo and perfectly complementary to Minish ;)

Maybe you're interested in adding SMMR, a more efficient alternative to MMR.

Paper: https://dl.acm.org/doi/10.1145/3726302.3730250
GitHub: https://github.com/kliakhnovich/smmr

On the other hand there is already a production-ready implementation provided in the repo.

[Pringled]

@do-me thanks! Great suggestion, I didn't come across this repo while looking for existing packages. I'll have a look; the dependency tree in the linked repo is very heavy and there's no tests etc, I'll see if I can add a numpy only tested version to pyversity, I think that's worth having!

[Pringled]

Hey @do-me, I just went through the paper and implementation, but unfortunately I think it doesn't add too much over regular MMR. The results are a bit mixed (very minor improvements), and it adds a lot of complexity to the API since you have 3 tunable parameters instead of 1 like with regular MMR. I'll try out the implementation though on some data and see how well it works; if it turns out to work very differently from MMR I might still add it later.

The positive: I will most likely add Sliding Spectrum Decomposition (SSD) (a baseline in the paper you linked), from this paper originally https://arxiv.org/pdf/2107.05204. This looks like a nice addition that would fit in well with the current API, and uses a very different approach from the strategies already supported in Pyversity.