The cpca package approaches Common Principal Component Analysis (CPCA) using the stepwise method proposed by Trendafilov. In contrast to others, this method orders the components by the explained variance intrincically and allows computing a few first components. The later feature is beneficial in practice for high-dimensional data.
The figure above shows an application of CPCA to the image compression problem. The original figure is the famous one from http://lenna.org, and we estimated CPCA for three RBG data matrices simultaneously rather than doing SVD/PCA for each matrix separately. (Spoiler: the performance of PCA and CPCA is visually the same, and this example is sought for the demonstration purpose only.)
Left panel on the Figure shows the CPCA-based compression using 5 components, central panel - 25 components, and right panel - all components. Script: learn/03-pca/R/05-compress-image-cpca.R.
The main function in the official release is cpc, while a new function cpca is used in the development branch.
library(cpca)
demo(iris, package = "cpca")
The demo.html shows that
the eigenvectors obtained by the cpc function are exactly the same as reported
in Trendafilov et al., 2010, Section 5, Example 2.
The following commands install the development (master branch) version from Github.
library(devtools)
install_github("cpca", user = "variani")
Currently, we don't have a specific publication for the cpca package. Please see the current citation information by the following command in R.
library(cpca)
citation(package = "cpca")
A list of publications where the cpca package was used:
- Kanaan-Izquierdo et al., Multiview approach to spectral clustering, EMBC, 2012
- Fernandez-Albert et al., Intensity drift removal in LC/MS metabolomics by common variance compensation, Bioinformatics, 2014
The cpca package is licensed under the GPLv3. See COPYING file in the inst directory for additional details.
- COPYING - cpca package license (GPLv3)