[R] caret package for time series applications

Max Kuhn mxkuhn at gmail.com
Thu Oct 1 22:58:09 CEST 2009


> Hello, I have some time series applications, where i have a large set of X
> variables (hundreds) and thousands of time data points (sampling every
> minute).
> I like to use the caret package to support the analysis, variable selection
> and model selection.  However, reading the documentation, it looks like
> caret uses resampling methods.  Not sure if these methods work with time
> series, as you need  a block bootstrap or such.
> My question:  Does the caret package support the usage of time series data?

I don't know much about resampling with time series. If regular
resampling methods are not appropriate, you can use the index argument
of trainControl to specify the indices of the samples that you want in
the training set. We can always add resampling options that are
specific to time series.

It also assumes that the predictors are in the form of a data frame or
matrix, so I'm not sure that passing an object of some other class
would work right now. I'd be happy to add new methods for train that
are specific to certain data types.



More information about the R-help mailing list