This repository contains the files related to the experiments reported in
A. Cristofari. Block cubic Newton with greedy selection. arXiv:2407.18150.
In the above paper, a second-order block coordinate descent method is proposed, named Inexact Block Cubic Newton (IBCN) method, using a greedy rule for the block selection and cubic models for the block update.
Andrea Cristofari (e-mail: andrea.cristofari@uniroma2.it)
IBCN is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. IBCN is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with IBCN. If not, see http://www.gnu.org/licenses/.
All codes are in Matlab. Two classes of unconstrained problems are considered, as described in the above paper.
-
For sparse least squares (non-convex problems), just run the file
main_sp_ls.m
. -
For l2-regularized logistic regression (convex problems), first download the datasets
gisette
,leu
andmadelon
from https://www.csie.ntu.edu.tw/~cjlin/libsvmtools/datasets/. Using thelibsvmread
software, which can be downloaded from therein as well, convert the files into matlab files and save them asgisette.mat
,leu.mat
andmadelon.mat
, respectively. In each matlab file, the instance matrix must be a sparse matrix namedA
and the label vector must be a vector namedb
. Then, run the filemain_l2_log_reg.m
.