Logistic Biplot by Conjugate Gradient Algorithms and Iterated SVD

Multivariate binary data are increasingly frequent in practice. Although some adaptations of principal component analysis used to reduce dimensionality for this kind data, none them provide a simultaneous representation rows and columns (biplot). Recently, technique named logistic biplot (LB) has been developed represent the matrix simultaneously, even though algorithm fit parameters is too computationally demanding be useful presence sparsity or when large. We propose fitting an LB model using nonlinear conjugate gradient (CG) majorization–minimization (MM) algorithms, cross-validation procedure introduced select hyperparameter that represents number dimensions model. A Monte Carlo study considers scenarios with several levels different set shows based on successful selection all algorithms studied. The comparison running times CG more efficient not very large, while performance MM better balanced As complement proposed methods give practical support, package written R language called BiplotML. To complete study, real gene expression methylation illustrate methods.