[MINIMUM,FVAL,EXITFLAG,OUTPUT] = FMINGRADRAND(FUN,PARS,[OPTIONS],[CONSTRAINTS], ...) random gradient optimizer This minimization method uses a gradient method with random directions. Namely, it first determine a random direction in the optimization space and then uses a Newton method. This is repeated iteratively until success. This method is both fast and less sensitive to local minima traps. The objective function has syntax: criteria = objective(p) Calling: fmingradrand(fun, pars) asks to minimize the 'fun' objective function with starting parameters 'pars' (vector) fmingradrand(fun, pars, options) same as above, with customized options (optimset) fmingradrand(fun, pars, options, fixed) is used to fix some of the parameters. The 'fixed' vector is then 0 for free parameters, and 1 otherwise. fmingradrand(fun, pars, options, lb, ub) is used to set the minimal and maximal parameter bounds, as vectors. fmingradrand(fun, pars, options, constraints) where constraints is a structure (see below). fmingradrand(problem) where problem is a structure with fields problem.objective: function to minimize problem.x0: starting parameter values problem.options: optimizer options (see below) problem.constraints: optimization constraints fmingradrand(..., args, ...) sends additional arguments to the objective function criteria = FUN(pars, args, ...) Example: banana = @(x)100*(x(2)-x(1)^2)^2+(1-x(1))^2; [x,fval] = fmingradrand(banana,[-1.2, 1]) Input: FUN is the function to minimize (handle or string): criteria = FUN(PARS) It needs to return a single value or vector. PARS is a vector with initial guess parameters. You must input an initial guess. PARS can also be given as a single-level structure. OPTIONS is a structure with settings for the optimizer, compliant with optimset. Default options may be obtained with o=fmingradrand('defaults') options.MinFunEvals sets the minimum number of function evaluations to reach An empty OPTIONS sets the default configuration. CONSTRAINTS may be specified as a structure constraints.min= vector of minimal values for parameters constraints.max= vector of maximal values for parameters constraints.fixed= vector having 0 where parameters are free, 1 otherwise constraints.step= vector of maximal parameter changes per iteration constraints.eval= expression making use of 'p', 'constraints', and 'options' and returning modified 'p' or function handle p=@constraints.eval(p) An empty CONSTRAINTS sets no constraints. Additional arguments are sent to the objective function. Output: MINIMUM is the solution which generated the smallest encountered value when input into FUN. FVAL is the value of the FUN function evaluated at MINIMUM. EXITFLAG return state of the optimizer OUTPUT additional information returned as a structure. Reference: Computer Methods in Applied Mechanics & Engg, Vol 19, (1979) 99 Contrib: Sheela V. Belur(sbelur@csc.com) 1998 [ossrs] Version: Nov. 26, 2018 See also: fminsearch, optimset (c) E.Farhi, ILL. License: EUPL.

Generated on Mon 26-Nov-2018 15:08:42 by