computeRawError {cvwrapr}

Compute the nobs by nlambda matrix of errors

Description

Computes the nobs by nlambda matrix of errors corresponding to the error measure provided. Only works for "gaussian" and "poisson" families right now.

Usage

computeRawError(predmat, y, type.measure, family, weights, foldid, grouped)

Arguments

predmat

Array of predictions. If 'y' is univariate, this has dimensions 'c(nobs, nlambda)'. If 'y' is multivariate with 'nc' levels/columns (e.g. for 'family = "multionmial"' or 'family = "mgaussian"'), this has dimensions 'c(nobs, nc, nlambda)'. Note that these should be on the same scale as 'y' (unlike in the glmnet package where it is the linear predictor).

y

Response variable.

type.measure

Loss function to use for cross-validation. See 'availableTypeMeasures()' for possible values for 'type.measure'. Note that the package does not check if the user-specified measure is appropriate for the family.

family

Model family; used to determine the correct loss function.

weights

Observation weights.

foldid

Vector of values identifying which fold each observation is in.

grouped

Experimental argument; see 'kfoldcv()' documentation for details.

Value

A list with the following elements:

cvraw

An nobs by nlambda matrix of raw error values.

weights

Observation weights.

N

A vector of length nlambda representing the number of non-NA predictions associated with each lambda value.

type.measure

Loss function used for CV.