DPS

DPS Correntropy based, hierarchical density preserving data split

    R = DPS(A,LEVELS,CLASSWISE)
    [R H] = DPS(A,LEVELS,CLASSWISE)

Input
A	Input dataset (PRDATASET)
LEVELS	Number of split levels, default: 3
CLASSWISE	Use (1, default) or ignore (0) label information

Output
R	Index array with rotation set with 2^LEVELS folds
H	Hierarchy of splits

Description

Density Preserving Sampling (DPS) divides the input dataset into a given  number of folds (2^LEVELS) by maximizing the correntropy between the folds  and can be used as an alternative for cross-validation. The procedure is  deterministic, so unlike cross-validation it does not need to be repeated.

Reference(s)

M. Budka, B. Gabrys, Correntropy-based density-preserving data sampling as an alternative to standard cross-validation, IJCNN2010, 1-8 http://www.budka.co.uk/

See also

datasets, crossval,