int main(int argc, char **argv)
{
int n;
int m;
minlbfgsstate state;
minlbfgsreport rep;
ap::real_1d_array s;
double x;
double y;
//
// Function minimized:
// F = exp(x-1) + exp(1-x) + (y-x)^2
// N = 2 - task dimension
// M = 1 - build tank-1 model
//
n = 2;
m = 1;
s.setlength(2);
s(0) = 10;
s(1) = ap::randomreal()-0.5;
minlbfgscreate(n, m, s, state);
minlbfgssetcond(state, 0.0, 0.0, 0.0001, 0);
minlbfgssetxrep(state, true);
printf("\n\nF = exp(x-1) + exp(1-x) + (y-x)^2\n");
printf("OPTIMIZATION STARTED\n");
while(minlbfgsiteration(state))
{
if( state.needfg )
{
x = state.x(0);
y = state.x(1);
state.f = exp(x-1)+exp(1-x)+ap::sqr(y-x);
state.g(0) = exp(x-1)-exp(1-x)+2*(x-y);
state.g(1) = 2*(y-x);
}
if( state.xupdated )
{
printf(" F(%8.5lf,%8.5lf)=%0.5lf\n",
double(state.x(0)),
double(state.x(1)),
double(state.f));
}
}
printf("OPTIMIZATION STOPPED\n");
minlbfgsresults(state, s, rep);
//
// output results
//
printf("X = %4.2lf (should be 1.00)\n",
double(s(0)));
printf("Y = %4.2lf (should be 1.00)\n\n\n",
double(s(1)));
return 0;
}
void OptHOM::OptimPass(alglib::real_1d_array &x, int itMax)
{
static const double EPSG = 0.;
static const double EPSF = 0.;
static const double EPSX = 0.;
static int OPTMETHOD = 1;
Msg::Info("--- Optimization pass with initial jac. range (%g, %g), jacBar = %g",
minJac, maxJac, jacBar);
iter = 0;
alglib::real_1d_array scale;
calcScale(scale);
int iterationscount = 0, nfev = 0, terminationtype = -1;
if (OPTMETHOD == 1) {
alglib::mincgstate state;
alglib::mincgreport rep;
try{
mincgcreate(x, state);
mincgsetscale(state,scale);
mincgsetprecscale(state);
mincgsetcond(state, EPSG, EPSF, EPSX, itMax);
mincgsetxrep(state, true);
alglib::mincgoptimize(state, evalObjGradFunc, printProgressFunc, this);
mincgresults(state, x, rep);
}
catch(alglib::ap_error e){
Msg::Error("%s", e.msg.c_str());
}
iterationscount = rep.iterationscount;
nfev = rep.nfev;
terminationtype = rep.terminationtype;
}
else {
alglib::minlbfgsstate state;
alglib::minlbfgsreport rep;
try{
minlbfgscreate(3, x, state);
minlbfgssetscale(state,scale);
minlbfgssetprecscale(state);
minlbfgssetcond(state, EPSG, EPSF, EPSX, itMax);
minlbfgssetxrep(state, true);
alglib::minlbfgsoptimize(state, evalObjGradFunc, printProgressFunc, this);
minlbfgsresults(state, x, rep);
}
catch(alglib::ap_error e){
Msg::Error("%s", e.msg.c_str());
}
iterationscount = rep.iterationscount;
nfev = rep.nfev;
terminationtype = rep.terminationtype;
}
Msg::Info("Optimization finalized after %d iterations (%d function evaluations),",
iterationscount, nfev);
switch(int(terminationtype)) {
case 1: Msg::Info("because relative function improvement is no more than EpsF"); break;
case 2: Msg::Info("because relative step is no more than EpsX"); break;
case 4: Msg::Info("because gradient norm is no more than EpsG"); break;
case 5: Msg::Info("because the maximum number of steps was taken"); break;
default: Msg::Info("with code %d", int(terminationtype)); break;
}
}
