This project is a work in progress
This is a port of NIPALS SCCA package written by Lee et. al. [1, 2] (scca). It is written for speedup in Rcpp.
Currently, only the lasso penalty function is implemented. Also, the only cross-validation method implemented is the sequential one-dimensional search described in the paper.
Make sure you have the devtools package installed as well as Rcpp and
RcppArmadillo:
install.packages(c("devtools", "Rcpp", "RcppArmadillo"))
library(devtools)You can install the current development version from github using devtools:
install_github("pimentel/fscca")The build sets the flag PKG_CXXFLAGS = "-std=c++11" in src/Makevars. This
is standard for g++ and clang. If this fails on your computer, let me know
(my guess it will likely fail on windows).
The branch master should always be stable. Please file a bug report if it is
unstable.
Assuming all goes well, load it like a usual package:
library(fscca)The main R entry point is the function fscca(). It is reasonably well
documented in R which you can access by typing ?fscca or help('fscca').
This function will automatically perform cross-validation as well as compute
multiple components (default is just the first).
If you don't want to perform cross-validation, you can compute the first
canonical vectors by using nipals_sparse() which is also documented in R.
- Implement other penalty functions (soft-thresholding, SCAD, HL)
- Three matrix extension (
fscca(X, Y, Z)) - Implement full grid search
Please report any bugs to Github: http://github.com/pimentel/fast-scca/issues