Unused parameters in dask_ml.cluster.KMeans

The precompute_distances, copy_x, and n_jobs parameters for the KMeans class do not seem to be used beyond initialization.

Are there plans for use of these parameters?

[TomAugspurger]

These are included solely to match the scikit-learn API

copy_x dosen’t make sense because dask arrays are immutable.

n_jobs doesn’t really make sense. In scikit-learn, that controls the parallelism of the n_init different random initilizations. We only do 1 initialization.

I’m not sure about precompute_distances. I suspect it makes less sense for Dask-ML, since the distances would have to move between workers, but I don’t recall the scikit-learn implementation well.

[TomAugspurger]

I think my general preference is to match the signature, but raise an exception when the value doesn’t match the default. That makes it a bit easier to do prototyping, since you can more easily get_params() and set_params() between a scikit-learn model and a dask-ml model.