k-fold cross validation - how to get the prediction automatically?

Question

This may be a silly question but I just can\'t find a package to do that...I know I can write some codes to get what I want but it would be nice to have a function to do it auto

Answer 1

As stated in the comments, caret makes cross-validation very easy. Just use the "glm" method, like so:

> library(caret)
> set.seed(2)
> dat <- data.frame(label=round(runif(100,0,5)),v1=rnorm(100),v2=rnorm(100))
> tc <- trainControl("cv",10,savePred=T)
> (fit <- train(label~.,data=dat,method="glm",trControl=tc,family=poisson(link = "log")))
100 samples
  2 predictors

No pre-processing
Resampling: Cross-Validation (10 fold) 

Summary of sample sizes: 90, 91, 91, 90, 90, 89, ... 

Resampling results

  RMSE  Rsquared  RMSE SD  Rsquared SD
  1.53  0.146     0.131    0.235      


> fit$finalModel$family

Family: poisson 
Link function: log 

> head(fit$pred)
      pred obs rowIndex .parameter Resample
1 2.684367   1        1       none   Fold01
2 2.165246   1       18       none   Fold01
3 2.716165   3       35       none   Fold01
4 2.514789   3       36       none   Fold01
5 2.249137   5       47       none   Fold01
6 2.328514   2       48       none   Fold01

Answer 2

I would suggest investigating cv.glm from package boot, because you are working with a glm model. Another option would be package cvTools. I've found it more useful to write up my own function for CV, though. It sounds like you want a CV function that ends halfway, and most CV functions I've seen will average the prediction error over all the validation sets and return just the average (which, of course, is the definition of cross validation).