multiTaskElasticNet: The multi-response elastic net (EN) function with multiple...

In CBIIT/rcellminerElasticNet: rcellminerElasticNet

Description

Applies the elastic net regression algorithm to learn a sparse linear model for predicting a response matrix from a set of input feature vectors.

Usage

multiTaskElasticNet(
  X,
  Y,
  alphaVals = seq(0.2, 1, length = 9),
  lambdaVals = NULL,
  nFoldsForParamSelection = 4,
  nTrainingRuns = 200,
  minFeatureFrequencyPctl = 0.95,
  verbose = TRUE,
  useLambda1se = TRUE,
  id = ""
)

Arguments

X	n x p matrix with input feature vectors along columns.
Y	n x k matrix of responses to be predicted using sparse linear combinations of input feature vectors specified in X
alphaVals	a vector of alpha values to be optimized over
lambdaVals	a vector of lambda values to be optimized over
nFoldsForParamSelection	the number of cross-validation folds to perform
nTrainingRuns	number of training runs to perform
minFeatureFrequencyPctl	a fractional value (0-1). Features in the x percentile defined by this parameter after the removal of features with zero weights are retained.
verbose	a boolean, whether debugging information should be displayed
useLambda1se	a boolean, whether to use the largest value of lambda such that the error is within 1 standard error of the minimum (lambda.1se) as the optimum lambda; if FALSE then lambda.min is used.
id	a optional string identifier for the EN run

Value

a list of enResults objects, each with members:

• predictorWts weights for selected predictors

• predictorSelectionFreq the frequency of the selected predictors across EN iterations

• predictedResponse cumulative predicted response across the predictors

• predictedResponseCor a vector whose nth element gives the correlation of the response with the linear combination of the first n predictors, with coefficient weights given in predictorWts.

• call the command used to call the function with parameters described

• alpha the optimized alpha value used

• lambda the optimized lambda value used

• foldIds the cross-validation IDs used

• cvm mean cross-validated errors for alpha values tried

• cvOutput an object containing all cross-validation information for the optimized alpha

• featureWtMat a matrix with the feature weights for all features across all training runs

Author(s)

Vinodh Rajapakse

CBIIT/rcellminerElasticNet documentation built on Sept. 8, 2020, 6:21 p.m.