Memory Layout

Most machine learning libraries expect row-major data where rows are observations and columns are features. Column-major internal storage keeps each column as a contiguous array. The custom DataFrame is internally column-major, so conversion is needed before passing data to a model.