I have a large p (~20K) and small n (~500) problem. The first thing I was thinking is dimension reduction. After trying PCA, robust PCA, ICA, removing highly correlated features, I was thinking to use Random Projection. However, there is no simple R implementation of Random Projection.
I have found a few random projection R packages, like
I concur that the
RPEnsemble package doesn't seem to expose the low-level methods that would allow you to use only that feature in any convenient form.
I did however come across this R source code which seems fairly straight-forward and reasonably documented: R source code for random projections. This is part of the clusterv package and you can download it there.