M1QN3

M1QN3#

Description#

M1QN3 is a quasi-Newtonian optimization algorithm originally developed at INRIA to minimize functions with a very high number of variables. The algorithm is described in Gilbert and Lemaréchal, 1989. The original code was written in Fortran; F. Chevallier translated it to Python in 2005.

The algorithm requires the following arguments:

  • niter: maximum number of iterations

  • nsim: maximum number of simulations

  • maxiter: maximum number of iterations; if one of the two previous

    is missing,:bash:niter is set to:bash:maxiter and:bash:nsim to 2 times maxiter

  • dxmin: absolute precision on x; optional; default is 1e-20

  • df1: expected decrease for f; optional; default is 0.01

  • epsg: relative precision on the gradient; optional; default is

    1e-20

  • mode: mode more M1QN3; optional; default is 0; for expert users

    only

  • m: number of updates; optional; default is 5; for expert users

    only

A Yaml template presents as follows:

minimizer :
  plugin:
    name: M1QN3
    version: std
  simulator:
    plugin:
      name: gausscost
      version: std
  maxiter: 10
  epsg: 0.03
  df1: 0.01

Requirements#

M1QN3 requires the following plugins to be executed properly: 1. a simulator to compute function to minimize; optional: default is (gausscost,:bash:std)

Contents