Splitting one dataset into training and test is still internal validation,
and it requires an enormous sample size (around 20,000 typically) in order
to be competitive with the bootstrap.
Note that the Design package has been replaced with rms, and rms has two
functions for external validation (val.prob and val.surv, should you have
had external data unlike your case) and many more for internal validation.
Frank
beginner wrote> I would like to do external validation using R software. So far I have
> used packages like "Design" and "DAAG". However they
perform internal
> validation rather than external one. In order to perform external
> validation I would have to split my data beforehand into training and test
> set, leave the test set on the site and use only the training set to
> select a model. I would then test the selected model with the sample set
> left initially on the site. I would like to repeat this process several
> times to make sure that all the samples are included at least once in a
> test set. I thought that I need to use a loop function in R to perform
> this process automatically. As I am new to R I don't know how to make a
> loop. Could you please help me with this or suggest an R package ? I would
> be very very grateful for help with this task !
-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context:
http://r.789695.n4.nabble.com/how-to-do-an-external-validation-with-R-tp4667404p4667426.html
Sent from the R help mailing list archive at Nabble.com.