Often, variables are linked to each other via a network. When such a network structure is known, this knowledge can be incorporated into regularized regression settings. In particular, an additional network penalty can be added on top of another penalty term, such as a Lasso penalty. However, when the type of interaction via the network is unknown (that is, whether connections are of an activating or a repressing type), the connection signs have to be estimated simultaneously with the covariate coefficients. This can be done with an algorithm iterating a connection sign estimation step and a covariate coefficient estimation step. We show detailed simulation results of such an algorithm. The algorithm performs well in a variety of settings. We also briefly describe the R-package that we developed for this purpose, which is publicly available.
The views expressed are those of the author(s) and do not necessarily represent those of the Bank of Lithuania.