Column major dataset abstraction for efficiently manipulating
in memory datasets.

Dealing with categorical dataset data involves having two mapping systems.
The first is a map of category to integer within the same column.
The second is a 'one-hot' encoding where you generate more columns but those have
a reduced number of possible categories, usually one categorical value per
column.

The etl pipeline and dataset operators are built to produce a metadata options map.
Their API access to the options is centralized in this file.

PCA and K-PCA using smile implementations.

A set of common 'pipeline' operations you probably will want to run on a dataset.

Helper functions for dealing with a sequence of maps.

Conversion mechanisms from dataset to tensor and back