Make Predictions
predict.transreg.Rd
Predicts outcome
Usage
# S3 method for transreg
predict(object, newx, stack = NULL, ...)
Arguments
- object
object of class
transreg
- newx
features: matrix with \(n\) rows (samples) and \(p\) columns (variables)
- stack
character "sta" (standard stacking) or "sim" (simultaneous stacking)
- ...
(not applicable)
Value
Returns predicted values or predicted probabilities. The output is a column vector with one entry for each sample.
References
Armin Rauschenberger, Zied Landoulsi, Mark A. van de Wiel, and Enrico Glaab (2023). "Penalised regression with multiple sets of prior effects". Bioinformatics (In press). doi:10.1093/bioinformatics/btad680 armin.rauschenberger@uni.lu
Examples
#--- simulation ---
set.seed(1)
n0 <- 100; n1 <- 10000; n <- n0 + n1; p <- 500
X <- matrix(rnorm(n=n*p),nrow=n,ncol=p)
beta <- rnorm(p)
prior <- beta + rnorm(p)
y <- X %*% beta
#--- train-test split ---
foldid <- rep(c(0,1),times=c(n0,n1))
y0 <- y[foldid==0]
X0 <- X[foldid==0,]
y1 <- y[foldid==1]
X1 <- X[foldid==1,]
#--- glmnet (without prior effects) ---
object <- glmnet::cv.glmnet(y=y0,x=X0)
y_hat <- predict(object,newx=X1,s="lambda.min")
mean((y1-y_hat)^2)
#> [1] 493.8406
#--- transreg (with prior effects) ---
object <- transreg(y=y0,X=X0,prior=prior)
y_hat <- predict(object,newx=X1)
mean((y1-y_hat)^2) # decrease in MSE?
#> [1] 265.7714