[R] plotting ROC from coefficient in the model

[David Winsemius]

On Jun 23, 2009, at 10:25 AM, sunny vic wrote:

> Hi everyone.
>    Probably this is statistical question rather than an R, but it  
> involves
> packages from R I am asking here since I am unable to find an  
> answer. In the
> parametric modeling packages like glmnet, lasso etc......., we are  
> able to
> obtain the coeffcients that have entered the model.
>
> for eg in glmnet if we are working on a dataset containing 15  
> variables
> the coeffecient parameters output is like this, from the below  
> result we
> know that 5 variables or features have entered the model and are  
> chosen and
> the rest 10 variables have not entered, can we plot an ROC curve  
> detremine
> sensitivity, specificity and confusion matrix using just this below
> information. any input would be great.
>
> 0.000
> 0.01213
> -0.1213
> 0.0000
> 0.0000
> 0.0000
> 0.0000
> -0.00034
> 0.0000
> 0.0000
> 0.0000
> 0.0000
> 0.0023
> 0.0988
> 0.0000

No. ROC's require a more complete picture of the data and model  
predictions than just the model coefficients. Because that was not  
clear, I worry about what else you may be doing with those packages.  
Besides the warnings about GIGO, it's also possible to make garbage  
out of perfectly good inputs by applying faulty methods. In either  
case, an intact sense of smell is essential.

-- 
David Winsemius, MD
Heritage Laboratories
West Hartford, CT