Supplementary files for "Constraining classifiers in molecular analysis: invariance and robustness"
ReadMe file for the supplementary material.
This provides the code used for training the (invariant) SVMs and predicting the class labels of a new matrix in R.
The code is bundled as the R package invariantSVM_1.0.tar.gz. It can be installed from command line by using the command:
R CMD INSTALL invariantSVM_1.0.tar.gz
After the package is loaded into the workspace (by using library(invariantSVM)), automatic parameter tuning in a multi-objective setting can be performed as shown in the following example:
#calculate 10x10 Accuracy on iris dataset over all invariances and R = 2library(TunePareto)print(tunePareto(data = as.matrix(iris[1:100, -ncol(iris)]),labels = iris[1:100,ncol(iris)],classifier = tunePareto.invariantSVM(), C=1, R = 2,invariance = c("lin","off","con","off&con","mon"),objectiveFunctions = list(cvAccuracy(10, 10))))
Collection of the analysed artificial datasets.
Collection of the fold lists used for the 10x10 cross-validation experiments.
This file provides the results of the experiments on artificial and transcriptome datasets. The accuracies and the selected number of features (R1-SVM) are given.
Our paper "Assessing phenotype order in molecular data" has been published in Scientific Reports.
Our paper "Representing dynamic biological networks with multi-scale probabilistic models"has been published in Communications Biology.