Implementation of the block coordinate descent procedure for solving the proximal function of latent group Lasso, highlighted by decomposing a DAG into several non-overlapping path graphs, and getting closed-form solution for each path graph. The procedure was introduced as Algorithm 4 in Yan and Bien (2015) <https://arxiv.org/abs/1512.01631> "Hierarchical Sparse Modeling: A Choice of Two Regularizers", and the closed-form solution for each path graph is solved in Algorithm 3 of that paper.
|Author||Xiaohan Yan <firstname.lastname@example.org>, Jacob Bien|
|Date of publication||2016-06-09 07:56:49|
|Maintainer||Xiaohan Yan <email@example.com>|
ancestor.find: Find ancestor nodes for a node in DAG.
hsm: Solves proximal operator of latent group Lasso in...
hsm-package: Block coordinate descent based on path graphs for proximal...
hsm.path: Solves proximal operator of latent group Lasso over a grid of...
lam.max.hsm: Computes the smallest lam value such that beta = 0.
path.find: Find all path graphs originated from a given root.
paths: Generate 'assign' and 'w.assign'.