predict.logilasso {logilasso}R Documentation

Predicts the interaction vector beta of a loglinear interaction model.

Description

Predicts the interaction vector(s) beta of a loglinear model log(p)=X*beta, fitted either by logilasso or levelcv. If lambda is specified, this lambda is taken to predict the beta for this value of lambda. If no value for lambda is specified, then the optimal value calculated by cross-validation is taken for objects of class cvlogilasso. For objects of class logilasso where no cross-validation was performed, the whole solution path for all lambdas is returned.

Usage

## S3 method for class 'logilasso':
predict(object, lambda = NULL, ...)
## S3 method for class 'cvlogilasso':
predict(object, lambda = NULL, ...)
## S3 method for class 'levellogilasso':
predict(object, lambda = NULL, to.which.int =
NULL, ...)

Arguments

object Object of class levellogilasso, cvlogilasso or class logilasso.
lambda Value for the penalization parameter lambda, for which the corresponding beta should be calculated.
to.which.int The number of factors the model should be predicted for.
... Additional arguments to predict function.

Value

Is either an object of class predlogilasso if a value for lambda was specified or if the optimal lambda can was assessed by cross-validation. Otherwise, if no value for lambda was specified and at the same time cvfold was chosen to be 1 (no cross-validation) it is of class predlogispez. The difference between these two classes is described below.

beta A predicted value for beta if the object is of class predlogilasso. For the class predlogispez this is a matrix consisting of the columns beta for the whole solution path.
lambda The lambda(s) for which the beta(s) was/were calculated.
probs Probabilities according to the model probs=exp(X*beta)
nls Negative likelihood score. For details see http://stat.ethz.ch/~dahinden/Paper/BMC.pdf
betapath The whole solution originally calculated path. For objects of class predlogispez this equals beta.
lambdapath The lambdas corresponding to the value betapath.
losspath nls for the whole solution path.

See Also

logilasso

Examples

library(gRbase)
data(reinis)

fit <- logilasso(reinis,lambdainit=1,lambdamin=0.1)
pred1 <- predict(fit,lambda=0.5)
pred2 <- predict(fit)

fitcv <- logilasso(reinis,lambdainit=1,lambdamin=0.1,cvfold=3)
predcv1 <- predict(fitcv)

levellogi <- levelcv(reinis,lambdainit=1,lambdamin=0.1,to.which.int=3,cvfold=3)
predlevel <- predict(levellogi)

## Methods plot and graphmod exist for all predicted models
## Except for pred2, there is no graphmod method, because no
## lambda was specified

plot(predcv1)
graphmod(predcv1)

[Package logilasso version 0.1.0 Index]