cv_iEN: Optimizes an iEN model via K-fold cross validation gridsearch...

In Teculos/immunological-EN: immunological Elastic-Net model, implementation and optimization

Description

Optimizes an iEN model via K-fold cross validation gridsearch and returns out-of-sample predictions and the associated model meta data.

Usage

cv_iEN(X, Y, foldid, alphaGrid, phiGrid, nlambda = 100, lambdas = NULL,
  priors, ncores, eval = c("RMSE", "RSS", "wilcox", "ROCAUC", "spearman",
  "pearson"), family = c("binomial", "gaussian"), intercept = TRUE,
  standardize = TRUE, center = TRUE)

Arguments

X	Input matrix of dimensions nobs x nfeat where each row is an observation vector.
Y	Response variable. Is continuous vector for family = "gaussian" and categorical (with two levels) for family = "binomial".
foldid	Vector that identifies which observations belong to which fold during K-fold Cross-Validation. foldid must consist of at least three folds for optimization and model estimation to occur.
alphaGrid	Vector of alpha values for model optimization.
phiGrid	Vector of phi values for model optimization.
nlambda	Lambda values are generated dynamically during cross-validation to avoid any data leak. nlambda determines the number of lambda values to generate.
lambdas	Optional vector of static lambda values.
priors	Continuous values which indicates immune features (columns of X) that are consistent with known biology. Values vary between 0 (low consistency) to 1 (highly consistent) for each immune feature which create the column space of X.
ncores	Number of cores to use during parallel computing of iEN cross-validation results. For optimal use set ncores = length(alphaGrid) * length(phiGrid).
eval	For binomial models evaluations using Wilcoxon P-value and ROCAUC are provided whereas for Gaussian models RMSE, RSS, Pearson P-value, and Spearman P-value are available.
family	Type of regression model, currently only "Binomial" and "Gaussian" are supported
intercept	Indicator for inclusion of regresstion intercept (default=TRUE).
standardize	Indication for X variable standardization prior to model fitting (default=TRUE).
center	Indication for X variable centering during scaling (default=TRUE).

Value

An object of class "iEN" is returned, which is a class composed of results from the K-fold cross validation and meta data about the analysis. The returned information includes:

Out-of-sample predictions from the K-fold cross validation. Evaluation of the out-of-sample predictions as defined by the eval parameter. Coefficients for each out-of-sample regression model, betas. the optimal parameters (alpha, lambda, phi) calculated for each fold of the analysis.

Examples

data(test_data)

alphaGrid <- seq(0,1, length.out=2)
phiGrid <- exp(seq(log(1),log(10), length.out=2))
nlambda <- 3
ncores <- 2
eval <- "RSS" 
family <- "gaussian"
intercept <- TRUE
standardize <- TRUE
center <- TRUE

model <- cv_iEN(X, Y, foldid, alphaGrid, phiGrid, nlambda, NULL, priors, ncores, eval, family, intercept, standardize, center)

Teculos/immunological-EN documentation built on Sept. 23, 2020, 11:53 p.m.