This code implements the estimation of robust regression models using the lasso procedure. Robustness is handled by modelling the residuals as arising from a Student-t distribution with an appropriate degrees-of-freedom. The optimization is performed using the expectation-maximization algorithm.
Primary features of the code:
* Automatically produce a complete lasso regularization path for a given degrees-of-freedom
* Select amount of regularization, and the degrees-of-freedom using cross-validation or information criteria
To cite this toolbox:
Schmidt, D.F. and Makalic, E.
Robust Lasso Regression with Student-t Residuals
Lecture Notes in Artificial Intelligence, to appear, 2017