Add remaining static clustering models

[ablaom]

https://juliastats.org/Clustering.jl/stable/

@OkonSamuel

[Yuan-Ru-Lin]

Hi @ablaom Affinity Propagation is used in the study that I am working on, but its interface hasn't been implemented in this package yet. Is there anything I should pay attention to if I have time to implement it next week?

[ablaom]

Many thanks for the offer of help!

My only comment is that this looks like a Static model, as it cannot generalise to new data instances (like KMeans can), correct?

Assuming that is the case, you can use DBSCAN as a template. See here for the workflow you would be supporting.

[Yuan-Ru-Lin]

It turns out I have some time in the weekend so I have drafted a PR (#28).

As for your question, the model looks a lot like KMedoids so I used it as a template and even directly extended the existing MMI.predict so that it works with AffinityPropagation. (I only realized it's Unsupervised instead of Static after I started working on it so sorry it's the wrong topic to talk about the model!)

[ablaom]

Mmm. If it's a truly static model (no generalisation) you can't use KMedoids without some changes:

• model type is Static
• fit has not training data (all the work is done by predict, which does of course need data)

[Yuan-Ru-Lin]

Judging from the algorithm of the model, I don't think the model can generalize to new data instances. (One can determine which exemplar a datum is closest to, but that's not necessarily a reasonable classifier for out-of-sample data.) It's better to implement it as a Static model. I'll do it later.

[ablaom]

Great. Ping me when you are ready for review. No rush.

[ablaom]

This is now merged. For @ablaom to do:

• Tag new release of this pkg
• Update model registry
• Add to list of models at MLJ docs
• Add model descriptor at MLJ docs

cc @Yuan-Ru-Lin

[Yuan-Ru-Lin]

I have existing PRs addressing some of the unchecked boxes

One is JuliaAI/MLJModels.jl#574, the other is JuliaAI/MLJ.jl#1147.

EDIT: add a PR