Bootstrapped CIs of MSE for (G)AM model
Hello,
There were several errors with your code. The following works but with
the other CI types.
library(ISLR)
library(mgcv)
library(boot)
# function to obtain MSE
MSE <- function(data, indices, formula) {
d <- data[indices, ] # allows boot to select sample
fit <- gam(formula, data = d)
ypred <- predict(fit)
mean((d[["wage"]] - ypred)^2)
}
data(Wage)
# Make the results reproducible
set.seed(1234)
# bootstrapping with 1000 replications
results <- boot(data = Wage, statistic = MSE,
R = 1000, formula = wage ~ education + s(age, bs =
"ps") + year))
# get 95% confidence intervals
# type = "bca" is throwing an error
ci.type <- c("norm","basic", "stud", "perc")
boot.ci(results, type = ci.type)
Hope this helps,
Rui Barradas
?s 20:36 de 22/11/2018, varin sacha via R-help escreveu:
Dear R-experts,
I am trying to get the bootstrapped confidence intervals of Mean squared error (MSE) for a (G)AM model. I get an error message.
Here below the reproducible R code. Many thanks for your response.
####################
install.packages("ISLR")
library(ISLR)
install.packages("mgcv")
library(mgcv)
install.packages("boot")
library(boot)
#MSE calculation
n=dim(Wage)[1]
p=0.667
GAM1<-gam(wage ~education+s(age,bs="ps")+year,data=Wage)
sam=sample(1?:n,floor(p*n),replace=FALSE)
Training =Wage [sam,]
Testing = Wage [-sam,]
ypred=predict(GAM1,newdata=Testing)
y=Testing$wage
MSE = mean((y-ypred)^2)
# Bootstrap 95% CI for MSE
# function to obtain MSE
MSE <- function(formula, data, indices) {
??d <- data[indices,] # allows boot to select sample
??fit <- gam(formula, data=d)
??return(MSE(fit))
} # bootstrapping with 1000 replications
results <- boot(data=Wage, statistic=MSE,
?? R=1000, formula=gam(wage ~education+s(age,bs="ps")+year,data=Wage)
)
# get 95% confidence intervals
boot.ci(results, type="bca")
##########################
______________________________________________ R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.