initialWt_Vfold: Compute initial estimates for online learning algorithms

In benkeser/onlinesl:

Description

This function creates a list of otherArgs used by each online algorithm by calling learner.init for each learner in SL.library.

Usage

initialWt_Vfold(Y, X, SL.library, V = 5, lossFn, ensembleControl,
  initialHessian = FALSE)

Arguments

Y	The outcome vector, usually of length nl.
X	The data.frame of predictors, with nl rows
SL.library	The K length character vector of wrappers for candidate online learners.
V	The number of folds for V-fold cross validation, defaults to 5
lossFn	The loss function used by the super learner
ensembleControl	The list of controls for the ensemble method. The function uses ensembleControl$ensemblePredictFn to compute predictions on held out data to minimize the lossFn using solnp.
hessian	Boolean of whether to return the estimated Hessian at the solution of the solnp algorithm. Must be set to TRUE if ensembleControl$updateWtFn requires an estimate of the Hessian to update the weights (e.g., if updateWtFn corresponds to second-order stochastic gradient descent).

Value

dsl The name of the algorithm from SL.library that was the discrete super learner.

alpha A K-length vector of initial weight estimates for the super learner

otherWtArgs A list with entry hessian containing the estimated hessian (set to null if initialHessian = FALSE).

benkeser/onlinesl documentation built on May 12, 2019, 12:09 p.m.