CrossValidate Options

CrossValidate Function Options

Usage

CrossValidate(BASIS,
              Target,
              nFolds,
              foldId = 0,
              Epis = "no",
              prior = "gaussian",
              search = "global")

Arguments

BASIS

Sample matrix with rows corresponding to samples and columns corresponding to features.

Target

Dependent variable for each observation. Usually a single column or vector.

nFolds

Number of folds for n-fold cross-validation.

Epis

Epistasis - "yes" or "no" for including two-way interactions.

Note: If set to "yes", the program adds two-way interactions, which increases computational complexity. This will add more columns to BASIS

foldId

Randomly assign samples to different folds.

Note: This will likely not be changed as this is set to 0 by default.

prior

Model prior - "binomial" or "gaussian"

In other words, is the dependent variable binomially-distributed or pseudo-normal?

search

Search type - "global" or "local"

Note: If set to "global", the program will search through all 400 combinations of alpha and lambda. Otherwise, the "local" search is performed using the logic from the original EBEN package, which skips certain iterations based on error changes.

• The empirical Bayesian elastic net is fully parallelized when performing a "global" search, which may take longer than the serial EBEN package as more iterations may be computed. For the "local" search, only the cross-validation steps are parallelized whereas the hyperparameter sweeping is still serial.