Skip to content

gregdvd/exclusive-lasso

Repository files navigation

Exclusive Group Lasso for Structured Variable Selection

Libraries implementing the algorithms described in the paper "Exclusive Group Lasso for Structured Variable Selection", available here.

The files are listed below with a short description. Check Matlab help (e.g., help Subset) for more information.

  • Subset: Class that describes an exclusive group and implements the main functions acting on it.
  • proximal: Proximal operator for the exclusive norm.
  • fista: FISTA algorithm for minimization with the exclusive norm as regularizer.
  • activeset: Implementation of the active set algorithm for minimization with the squared exclusive norm as regularizer.
  • activestring: Implementation of the active set algorithm for minimization with the squared exclusive norm as regularizer. This is a modified version that looks for long strings of consecutive active entries.
  • solveRestrictedVar: Algorithm to solve the restricted minimization problem with the squared exclusive norm as regularizer -- exploiting variational formulation.
  • solveRestrictedIP: Algorithm to solve the restricted minimization problem with the squared exclusive norm as regularizer -- requires Matlab's Optimization Toolbox.
  • fistabasic: Classic FISTA algorithm for minimization with the norm-1 regularizer.
  • proximalOverlap: Proximal operator for the group lasso with overlap.
  • fistaOverlap: FISTA algorithm for minimization with the "group lasso with overlap" regularizer.
  • test: simple example script applying the above algorithms to the same support detection problem and comparing results.

About

Exclusive sparsity support recovery

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages