fps is an R package that provides an implementation of an ADMM algorithm for computing the
Fantope projection and selection estimator. Most of the package is written in
C++ using Rcpp and the Armadillo C++ library. The estimator is based on
a convex relaxation of the sparse PCA problem based on the convex hull
projection matrices (the Fantope). This circumvents orthogonality and
deflation issues that plague previous approaches to sparse PCA. A
preliminary report describing this estimator and some of its near-optimal
statistical properties estimator can found in the
NIPS conference proceedings. A longer report with more details and new
results is forthcoming and will be posted to arXiv.
Use devtools to install directly from GitHub:
library(devtools)
install_github("fps", "vqv")library(fps)
data(wine)
out <- fps(cor(wine), ndim = 2)
plot(out)
# Extract basis coefficients for a particular solution
v <- coef(out, lambda = 0.5)
print(v)or
library(fps)
example(fps)- This package is currently under development and has not been thoroughly tested.
- The documentation is under development and minimal.
- The author has only built the package on Mac OS X 10.9.