Add wrappers for clustering to get uniform interface

[juliohm]

Can you please provide an example with

1. Clustering.KMeans
2. Clustering.DBSCAN
3. BetaML.GaussianMixtureClusterer

where we call the same MLJ interface functions to "train" the clustering models with a random data set X1 = rand(3,100) and then "predict" the labels of new points X2 = rand(3, 10)?

Every time I try to accomplish this task I get stuck with various issues (e.g. the output of the models follow different formats and conventions). I am looking for a simple example with MLJ.fit, MLJ.predict, MLJ.predict_mode and MLJ.transform only. I don't want to wrap the models into a machine in this example.

[ablaom]

It is not possible at present as 1, 2, and 3 are all different: 2 does not generalize to new observations, and 3 outputs probabilistic labels. There are natural differences in the API for this reason. For a uniform interface, we need a wrapper to convert non-generalizing clusterers to generalizing ones, and a wrapper to convert probabilistic-predicting clusterers into deterministic ones. These do not exist at present. They could be created using MLJ learning networks.

[juliohm]

No longer needed.