Wrap GridSearchCV and RandomizedSearchCV for MLDataset (xarray data structures)

[PeterDSteinberg]

PR #192 added dask-searchcv (which I understand will be dask-ml soon) base classes to Elm - the base classes used in elm.model_selection.EaSearchCV. PR #192 improved the evo algorithm as well as xarray (xarray_filters.MLDataset) support (See also #221 in progress for cross validation and MLDataset issues). It should be easy to just take GridSearchCV and RandomizedSearchCV from dask-ml / dask-searchcv and use the sklearn mixins for xarray support (use EaSearchCV for guidance).

[PeterDSteinberg]

Hi @gbrener I assigned this one to you - seems like a good way to get familiar with newer Elm changes without too big of a file diff needed. This can happen post-AGU if needed.

[PeterDSteinberg]

These classes will need to implement the sampler logic of EaSearchCV (PR #221). Let's make:

• elm.model_selection.GridSearchCV - same as dask_searchcv.GridSearchCV but can use a sampler for elm.pipeline.Pipeline- and MLDataset-based parameter searches.
• elm.model_selection.RandomizedSearchCV - same as dask_searchcv.RandomizedSearchCV but can use a sampler for elm.pipeline.Pipeline- and MLDataset-based parameter searches.

Maybe some of what is now elm.model_selection.EaSearchCV can become a base class (inheriting from dask_searchcv.DaskSearchBase) and that base class can be a basis for:

• elm.model_selection.GridSearchCV
• elm.model_selection.RandomizedSearchCV
• elm.model_selection.EaSearchCV