In developing a machine learner to classify sentences in plain text
sources of scientific documents I have been using the caret
package and
following the procedures described in the vignettes. What I miss
in the
package -- but quite possibly I am overlooking it! -- is functions
that allow
me to estimate whether a model has too much variance or too much
bias. In
particular, I am looking for something that helps me compare the
training and
test errors as functions of training set size. Clearly, I can
program
something in R, but if there is already a function in some package
to do that,
I would prefer to use it, of course.
Regards,
Richard
[[alternative HTML version deleted]]