9
$\begingroup$

I've been experimenting with the rfe function in the caret package to do logistic regression with feature selection. I used the lmFuncs functions with the following rfeContol :

ctrl <- rfeControl(functions = lmFuncs, method = 'cv', rerank=TRUE, saveDetails=TRUE, verbose = TRUE, returnResamp = "all", number=100)

Below is the structure of the rfe call:

fit.rfe=rfe(df.preds,df.depend, metric='RMSE',sizes=c(5,10,15,20), rfeControl=ctrl)

df.preds is a data frame of inputs to the model. df.depend is a vector of 1 or 0 corresponding to each row in df.preds to indicate response.

The resulting model accessed in from the fit object in the rfe object is of class lm and produces predicted values of less than zero and greater than 1 when I use the following code with the predict function:

predict(fit.rfe$fit,df,type='response')

Given I'm expecting this to be a logistic, all predicted values should greater than zero and less than one.

Any help will be appreciated.

Improve this question
$\endgroup$
2
  • 2
    $\begingroup$ Using 100-fold cross-validation is not advisable. Practically speaking, each fold is going to be 99% similar to another fold. Use a smaller number of folds (like 5 or 10) to get more new data in each fold (mix it up more) and then if you want to do repeats, specify repeats=. Otherwise you won't really simulate how the technique performs on new data. See this question for comparison of k-fold and leave-one-out CV, which is what you approach as number= goes up $\endgroup$
    – C8H10N4O2
    Commented Apr 19, 2016 at 12:44
  • $\begingroup$ Has you try lrFunc instead of lmFunc? $\endgroup$
    – Nestor Montano
    Commented Apr 5, 2019 at 23:58

1 Answer 1

Reset to default
6
$\begingroup$

lmFuncs is fitting linear regression. Just type lmFuncs$fit to see. Try to rewrite it: 

lmFuncs$fit<-function (x, y, first, last, ...){   
     tmp <- as.data.frame(x)   
     tmp$y <- y   
 glm(y ~ ., data = tmp,family=binomial)   
}

Note, that I don't know how to attach <environment: namespace:caret> and what is its meaning. You may try this trick on your data and comment the result.

Improve this answer
$\endgroup$
4
  • $\begingroup$ Thank you for your response. I get an error when I try to rewrite. lmFuncs$fit<-function (x, y, first, last, ...){ tmp <- as.data.frame(x) tmp$y <- y glm(y ~ ., data = tmp,family=binomial) } Error: unexpected symbol in "lmFuncs$fit<-function (x, y, first, last, ...){ tmp <- as.data.frame(x) tmp" $\endgroup$
    – ansek
    Commented Sep 23, 2012 at 19:08
  • $\begingroup$ Original code with paragraphes was going fine. I have just edited my answer. $\endgroup$
    – O_Devinyak
    Commented Sep 23, 2012 at 19:16
  • 4
    $\begingroup$ I got it to work by defining a new object called glmFuncs using the following code: glmFuncs=lmFuncs glmFuncs$fit=function (x, y, first, last, ...) { tmp <- as.data.frame(x) tmp$y <- y glm(y ~ ., data = tmp, family=binomial(link='logit'))}. It is doing exactly what I want it to do. Thank you soooo much. $\endgroup$
    – ansek
    Commented Sep 23, 2012 at 19:35
  • $\begingroup$ It's telling me {: task 6 failed - "Results do not have equal lengths" $\endgroup$
    – nigelhenry
    Commented Mar 10, 2023 at 14:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.