A wrapper built around the libLBFGS optimization library by Naoaki Okazaki. The lbfgs package implements both the Limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) and the Orthant-Wise Quasi-Newton Limited-Memory (OWL-QN) optimization algorithms. The L-BFGS algorithm solves the problem of minimizing an objective, given its gradient, by iteratively computing approximations of the inverse Hessian matrix. The OWL-QN algorithm finds the optimum of an objective plus the L1-norm of the problem's parameters. The package offers a fast and memory-efficient implementation of these optimization routines, which is particularly suited for high-dimensional problems.
copied from cf-staging / r-lbfgs