[FEATURE]: Mahalanobis Distance Kennard-Stone (MDKS) Sampler

[JacksonBurns]

Is your feature request related to a problem? Please describe.

Better interpolative splits for Artificial Neural Networks in particular with Mahalanobis Distance Kennard-Stone here.

Use-cases/examples of this new feature

See linked paper for specific examples, but is reported to generally provide data splits for ANN applications.

Desired solution/workflow

Using the base implementation of the Kennard-Stone sampler, implement this.

Discussion

Unforunately without re-writing the Mahalanobis distance method to accept pre-computed pairwise distances this method will scale at least O(n^3)

[JacksonBurns]

This can actually be achieved with something like this:

_, _, _, indexes_train, indexes_val, indexes_test = train_val_test_split(
    responses,
    train_size=0.8,
    val_size=0.1,
    test_size=0.1,
    sampler="kennard_stone",
    random_state=42,
    hopts={"metric": "mahalanobis"},
    return_indices=True,
)

...and it's not terribly slow.

[JacksonBurns]

I have confirmed that the Mahalanobis distance is working as intended, although the example above should use sampler="spxy" since (despite the name) MDKS is derived from SPXY (which is itself derived from Kennard Stone). We can just add a row in the README samplers table saying that you can achieve this sampling using SPXY.