[an error occurred while processing this directive] gsCCA Software
(aside image)

gsCCA: Bayesian CCA via group sparsity

NOTE: The contents of this page is not updated anymore. Instead, see CCAGFA .

The gsCCA code implements the Bayesian canonical correlation analysis model as presented in the paper Bayesian CCA via group sparsity by Seppo Virtanen, Arto Klami, and Samuel Kaski in International Conference on Machine Learning (ICML 2011).

The code can be used for solving the canonical correlation analysis (CCA) problem, and is particularly useful for scenarios with small sample counts and/or high dimensionalities. The code is the recommended way of computing Bayesian CCA. Besides implementing CCA, the model provides a more flexible latent variable representation separating the variation in two data sets into the correlating components and components describing within-data variation. The code also includes functionality for making predictions from one data to another, along the lines it was used for multilabel prediction in the paper.

This is experimental software provided as is; we welcome any comments and corrections but cannot give any guarantees about the code. If you have any comments or bug reports, please direct them to Seppo Virtanen.


The following files are included in the code release:

Installation and usage

The software package runs under R, a free language and environment for statistical computing. The package is tested with the R version 2.14.1 in the Linux environment, but should work with most other versions as well. No installation is required, just place the files in a directory where R can find them (for example, the working directory).

The software is released under the FreeBSD license. We kindly ask you to refer to the ICML paper when using the code.