For sparse learning problems, such as sparse generalized linear models and sparse undirected graphical model estimation, the current R packages still have a lot of room for improvement in terms of speed and robustness, especially in high dimensional settings or when there’s multi-colinearity among features in the data. We aim to build an R package that can achieve state-of-the-art performance on three model classes (1) sparse generalized linear model estimation, (2) sparse nonparametric generalized linear model estimation and (3) sparse undirected graphical estimation under a novel active-set based second-order optimization algorithmic framework. Statistical inference, such as post-regularization p-value computation, will also be implemented as an important feature in our package.


Jason Ge


  • Xingguo
  • Tuo Zhao