Software

• sgdGMF: An R/C++ package for the estimation of high-dimensional matrix factorization models under dispersion exponential family likelihoods, including quasi-likelihood estimating equations and Negative-Binomial models. In particular, the package implements well-established deterministic estimation algorithms, such as iterated least squares and quasi-Newton, as well as innovative stochastic optimization methods, such as adaptive stochastic gradient descent. It also provides several efficient initialization approaches, a rich environment of functions for postprocessing analysis and the possibility to perform parallel computing. The stable and development versions of the package are available, respectively, on CRAN (link) and GitHub (link).

• BayesGLMM.jl: A Julia package for the approximate estimation of Bayesian generalized linear mixed and additive models via efficient non-conjugate variational message passing. The package supports all the most commonly used exponential family likelihoods, as well as quasi- and pseudo-likelihood models, including non-regular loss functions in a generalized Bayes framework. Some examples are support vector machines, Huber regression, quantile regression and expectile regression. An R/C++ implementation of this software will be released soon.