Christine SINOQUET wrote:
Hello,
I would like to use the ROCR package to draw ROC curves and compute AUC
values.
However, in the specific context of my application, the true positive
rates and false positive rates are already provided by some upstream
method.
Of course, I can draw a ROC plot with the following command :
plot(x=FPrate, y=TPrate, "o", xlab="false positive rate", ylab="true
positive rate", xlim=c(0, 1), ylim=c(0, 1)
but this will bot compute the AUC value.
There are two possibilities :
Either it is possible to use the above parameters - FPrate and TPrate
vectors- to run the performance function and I would like to know how,
or it is not possible and I have to compute the area under the curve but
I cannot find on the Web how to perform this, through an R package,
using the two vectors above, if possible (I would rather not implement
an integration algorithm).
I thank you in advance for your answer.
My answer doesn't use R...
For some applications, when the classifier performance is quite good, integrating the ROC curve
isn't a good idea, as the curve is very abrupt.
Another approach is to use the fact that the (1-AUC) represents the probability of taking at random
one positive and one negative event and don't geting them in the correct order (by their score).
This is related to the Wilcoxon distribution.
So, if you have a set of negative and positive events and the "scores" assigned to them, you can
sort them by their scores and enumerate the number of couples out of order you can get.
Hope this help.
José-Marcio
--
---------------------------------------------------------------
Jose Marcio MARTINS DA CRUZ http://j-chkmail.ensmp.fr
Ecole des Mines de Paris
60, bd Saint Michel 75272 - PARIS CEDEX 06
mailto:jose-marcio.mart...@mines-paristech.fr
______________________________________________
R-help@r-project.org mailing list
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.
