This submission contains an interface to a LBFGS-code for unconstrained minimization problems of the form
min_{x} f(x) + c*|\tilde{x}|,
where x are the parameters, f is a scalar valued real function, c is a positive scalar (default value=0), |.| denotes the 1-norm, and \tilde{x} is some subset of the parameters x.
The submission is based on a C-port of a Fortran code by Jorge Nocedal, due to Naoaki Okazaki. In addition it also implements a method for 1-norm regularization referred to as "Orthant-Wise Limited-memory Quasi-Newton" due to Galen Andrew and Jianfeng Gao.
Note that in order to use this code, it is necessary to download and compile libLBFGS, currently available at http://www.chokkan.org/software/liblbfgs/.
Installation should be straightforward:
1. Download liblbfgs
2. Make necessary modifications in "make.m" such that the files "lbfgs.c" and "lbfgs.h" can be found by your compiler.
3. Run make.m in Matlab
(prior to running make.m, it might be necessary to run "mex -setup" in the Matlab prompt to locate installed, supported, compilers)
Comments and suggestions are greatly appreciated. |