ordBTL (version 0.8)

BTLboost: Boosting ordinal Bradley-Terry-Luce models

Description

Boosting procedure for ordinal BTL models

Usage

BTLboost(formula, data, objects=NULL, groupVars=NULL, selection=c("DEVIANCE","AIC","BIC"), mstop=500, nu=1, maxit=1, verbose=TRUE, ...)

Arguments

formula

a formula describing the full model.

data

a data frame containing the design matrix for the model (See also design to generate such an design matrix).

objects

(optional) a character vector specifying the objects that should always be part of the model.

groupVars

(optional) a character vector specifying the subject-specific covariates, whose subject-object interactions are considered simultaneously in each boosting step. By default (groupVars=NULL), all subject-object interactions are cosidered separately in each boosting step.

selection

a character specifying the criterion that is used in each boosting step to determine the best fitting covariate(s).

mstop

an integer giving the number of boosting iterations.

nu

a double between 0 and 1 defining the step size or shrinkage parameter.

maxit

an integer representing the maximum number of Fisher-scoring iterations (see also vglm.control).

verbose

logical indicating if output should be produced for each boosting iteration.

...

further arguments passed to ordBTL.

Value

A List of

• BEST contains estimated parameters of the last boosting iteration
• AIC a vector of AIC values for each boosting iteration
• BIC a vector of BIC values for each boosting iteration
• DEVIANCE a vector that reflects the deviance of each boosting iteration
• PATH a dataframe containing the coefficient build-up at the end of each boosting iteration
• UPDATED a vector of strings containing the selected components in each boosting iteration

Examples

# Get the CEMS data and generate design matrix
example(wide2long, package="ordBTL", echo=FALSE)

des2 <- design(CEMSlong[-which(is.na(CEMSlong$Y)),], 
               var1="object1", var2="object2", 
               use.vars="ALL", reference="Stockholm")

# Formula for full model considering all subject-object interactions
form2 <- Y ~ 
  (GAMMA.London+GAMMA.Paris+GAMMA.Milano+GAMMA.StGallen+GAMMA.Barcelona)+
  (GAMMA.London+GAMMA.Paris+GAMMA.Milano+GAMMA.StGallen+GAMMA.Barcelona):
  (WOR+SEX+DEG+STUD+ENG+FRA+SPA+ITA)

## Not run: 
# Exemplatory boosting call with mstop=5
#BoostDev <- BTLboost(form2, data=des2, groupVars=c("WOR","DEG","SEX","STUD"), 
#                     selection="DEVIANCE", mstop=5)
## End(Not run)