cvAll: Cross Validate all nested models

In kohleth/kmisc: Kohleth Miscellaneous

Description

It fits all subset regression, allowing interaction and categorical covariates, and cross validates them. The code borrows heavily from package meifly.

Usage

cvAll(x, ...)

## Default S3 method:
cvAll(x, data, k, lossfn = .calcMSE, ind, finalfit = TRUE)

## S3 method for class 'lm'
cvAll(x, data, k = 5, ind = NULL, ...)

## S3 method for class 'lme'
cvAll(x, data, k = 5, ind = NULL, ...)

## S3 method for class 'merMod'
cvAll(x, data, k = 5, ind = NULL, ...)

validateAll(x, data, method, valid_prob, ...)

Arguments

x	Either an lm object or a two side formula.
data	Data.
k	Postive integer. k-fold cross validation is carried out. Defaults to 5.
lossfn	A function with 2 arguments. The first takes an array of predicted value, the second takes a vector of observed value. Then it calculates the loss. It defaults to a function which calculates the mean sqaured error (MSE).
finalfit	if TRUE models are refitted to the entire dataset at the end.
valid_prob	Positive number between 0 and 1. It is the proportion of data used in the validation set.

Value

An maxtrix of predictions on the validation set, with the following attributes:

loss The loss calculated by lossfn.

data The data used.

models The fitted models or just formulae of models, depending on finalfit.

Examples

## Using cvAll

allreg=cvAll(conc~uptake*Treatment,CO2)
allreg
attr(allreg,'models') 

calcmae=function(pred,obs)mean(abs(pred-obs),na.rm=TRUE)
allreg=cvAll(conc~uptake*Treatment,CO2,lossfn=calcmae)

## Using cvAll with class lm
lm0=lm(conc~uptake*Treatment,CO2)
allreg2=cvAll(lm0,CO2)

kohleth/kmisc documentation built on May 20, 2019, 12:53 p.m.