https://github.com/cran/DESP
Tip revision: 2aa2ecc69f210a4d02ffb836d9ddc57508228582 authored by Samuel Balmand on 02 February 2017, 10:33:51 UTC
version 0.2-2
version 0.2-2
Tip revision: 2aa2ecc
README
This package contains code, help and examples for robust estimation of sparse precision-matrices, as described in "On estimation of the diagonal elements of a sparse precision matrix" by Samuel Balmand and Arnak S. Dalalyan (posted as http://arxiv.org/abs/1504.04696) and in "Convex programming approach to robust estimation of a multivariate Gaussian model" (posted as http://arxiv.org/abs/1512.04734).
Contributors:
Arnak Dalalyan <Arnak.Dalalyan@ensae.fr>
Samuel Balmand <Samuel.Balmand@ensg.eu>
---------------
This package contains the following methods:
DESP_MST Estimation of DESP using minimum spanning trees
DESP_PML Estimation of DESP by penalized likelihood minimization
DESP_RML Estimation of DESP by likelihood maximization
DESP_RV Estimation of DESP by residual variance
DESP_SPT Estimation of DESP using shortest path trees
DESP_SRL_B Estimation of the coefficient matrix
sqR_Lasso Solve the square-root LASSO
scsSOCP Solve a second-order cone program using SCS
desp Robust estimation of a sparse precision matrix
desp.cv Selection of the tuning parameters of desp by v-fold cross-validation
The functions not prefixed by "DESP_" would be most convenient for most of the uses.
---------------
Requirements:
The open source solver SCS is already available in this package (see https://github.com/cvxgrp/scs), but to use the convex optimization proprietary software MOSEK or GUROBI that have an R interface, one has to install them separately. We highlight that the use of MOSEK and GUROBI solvers should be considered as deprecated. Their support will be removed in future versions of this package.
For estimating the coefficient matrix with the square-root Lasso with the MOSEK solver, DESP needs a working version of the Rmosek package that requires MOSEK and a MOSEK license (see http://rmosek.r-forge.r-project.org/).
For estimating the coefficient matrix with the square-root Lasso with the GUROBI solver, DESP needs a working version of the gurobi package that requires GUROBI and a GUROBI license (see http://www.gurobi.com/documentation/6.0/refman/r_api_overview.html).
To make the best use of this package, we recommend to use a compiler that supports OpenMP (see http://openmp.org).