TaoTermCreateL1#

Create a TaoTerm for the objective function term \(\|x - p\|_1\).

Synopsis#

#include "petsctao.h" 
PetscErrorCode TaoTermCreateL1(MPI_Comm comm, PetscInt n, PetscInt N, PetscReal epsilon, TaoTerm *term)

Collective

Input Parameters#

comm - the MPI communicator where the term will be computed
n - the local size of the \(x\) and \(p\) vectors (or PETSC_DECIDE)
N - the global size of the \(x\) and \(p\) vectors (or PETSC_DECIDE)
epsilon - a non-negative smoothing parameter (see TaoTermL1SetEpsilon())

Output Parameter#

term - the TaoTerm

Note#

If you would like to add an L1 regularization term \(\alpha \|x\|_1\) to the objective function of a Tao, do the following:

  VecGetLocalSize(x, &n);
  VecGetSize(x, &N);
  TaoTermCreateL1(PetscObjectComm((PetscObject)x), n, N, 0.0, &term);
  TaoAddTerm(tao, "reg_", alpha, term, NULL, NULL);
  TaoTermDestroy(&term);

If you would like to have a dictionary matrix term \(\alpha \|D x\|_1\), do the same but pass D as the map of the term:

  MatGetLocalSize(D, &m, NULL);
  MatGetSize(D, &M, NULL);
  TaoTermCreateL1(PetscObjectComm((PetscObject)D), m, M, 0.0, &term);
  TaoAddTerm(tao, "reg_", alpha, term, NULL, D);
  TaoTermDestroy(&term);

Level#

beginner

Location#

src/tao/term/impls/l1/taoterml1.c

Examples#

src/tao/unconstrained/tutorials/elastic_net_regularization.c

Index of all TaoTerm routines
Table of Contents for all manual pages
Index of all manual pages