/* nag_glopt_nlp_pso (e05sbc) Example Program.
*
* Copyright 2017 Numerical Algorithms Group.
*
* Mark 26.1, 2017.
*/
#include <math.h>
#include <nag.h>
#include <nag_stdlib.h>
#include <nage05.h>
#include <nagx02.h>
#include <nagx04.h>
#ifdef __cplusplus
extern "C"
{
#endif
static void NAG_CALL objfun_schwefel(Integer *mode, Integer ndim,
const double x[], double *objf,
double vecout[], Integer nstate,
Nag_Comm *comm);
static void NAG_CALL confun(Integer *mode, Integer ncon, Integer ndim,
Integer tdcj, const Integer needc[],
const double x[], double c[], double cjac[],
Integer nstate, Nag_Comm *comm);
static void NAG_CALL monmod(Integer ndim, Integer ncon, Integer npar,
double x[], const double xb[], double fb,
const double cb[], const double xbest[],
const double fbest[], const double cbest[],
const Integer itt[], Nag_Comm *comm,
Integer *inform);
#ifdef __cplusplus
}
#endif
static Integer display_option(const char *optstr, const Integer iopts[],
const double opts[]);
static void display_result(Integer ndim, Integer ncon, const double xb[],
double fb, const double cb[], const Integer itt[],
Integer inform);
/* Global constants.*/
/* Set the behaviour of the monitoring function.*/
static const Integer detail_level = 0;
static const Integer report_freq = 100;
/* Known solution for a comparison.*/
static const double f_target = -731.70709230672696;
static const double c_scale[] = { 2490.0, 750000.0, 0.1 };
static const double c_target[] = { 0.0, 0.0, 0.0 };
static const double x_target[] = { -394.1470221120988, -433.48214189947606 };
int main(void)
{
/* This example program demonstrates how to use
* nag_glopt_nlp_pso (e05sbc) in standard execution, and with
* nag_opt_nlp (e04ucc) as a coupled local minimizer.
* The non-default option 'Repeatability = On' is used here, giving
* repeatable results.
*/
/* Scalars */
Integer ncon = 3, ndim = 2, npar = 20;
Integer exit_status = 0, lcvalue = 17;
Integer liopts = 100, lopts = 100;
double fb, rvalue;
Integer i, inform, ivalue;
/* Arrays */
static double ruser[3] = { -1.0, -1.0, -1.0 };
char cvalue[17], optstr[81];
double opts[100], *bl = 0, *bu = 0, *cb = 0, *xb = 0;
double *cbest = 0, *fbest = 0, *xbest = 0;
Integer iopts[100], itt[7];
/* Nag Types */
Nag_VariableType optype;
Nag_Comm comm;
NagError fail;
/* Print advisory information. */
printf("nag_glopt_nlp_pso (e05sbc) Example Program Results\n\n");
/* For communication with user-supplied functions: */
comm.user = ruser;
printf("Minimization of the Schwefel function.\n");
printf("Subject to one linear and two nonlinear constraints.\n\n");
/* Allocate memory for arrays. */
if (!(bl = NAG_ALLOC(ndim + ncon, double)) ||
!(bu = NAG_ALLOC(ndim + ncon, double)) ||
!(cb = NAG_ALLOC(ncon, double)) ||
!(cbest = NAG_ALLOC(ncon * npar, double)) ||
!(fbest = NAG_ALLOC(npar, double)) ||
!(xb = NAG_ALLOC(ndim, double)) ||
!(xbest = NAG_ALLOC(ndim * npar, double)))
{
printf("Allocation failure\n");
exit_status = -1;
goto END;
}
for (i = 0; i < npar; i++)
fbest[i] = 0.0;
for (i = 0; i < npar * ndim; i++)
xbest[i] = 0.0;
for (i = 0; i < npar * ncon; i++)
cbest[i] = 0.0;
/* Set problem specific values. */
/* Set box bounds. */
for (i = 0; i < ndim; i++) {
bl[i] = -500.0;
bu[i] = 500.0;
}
/* Set constraint bounds. */
bl[ndim] = -1.0e6;
bl[ndim + 1] = -1.0;
bl[ndim + 2] = -0.9;
bu[ndim] = 10.0;
bu[ndim + 1] = 5.0e5;
bu[ndim + 2] = 0.9;
/* Initialize NagError structure. */
INIT_FAIL(fail);
/* Initialize the option arrays for nag_glopt_nlp_pso (e05sbc)
* using nag_glopt_opt_set (e05zkc).
*/
nag_glopt_opt_set("Initialize = e05sbc", iopts, liopts, opts, lopts, &fail);
if (fail.code != NE_NOERROR) {
printf("Error from nag_glopt_opt_set (e05zkc).\n%s\n", fail.message);
exit_status = 1;
goto END;
}
/* Query some default option values. */
printf(" Default Option Queries:\n\n");
exit_status = display_option("Constraint Norm", iopts, opts);
if (exit_status)
goto END;
exit_status = display_option("Maximum Iterations Completed", iopts, opts);
if (exit_status)
goto END;
exit_status = display_option("Distance Tolerance", iopts, opts);
if (exit_status)
goto END;
/* ------------------------------------------------------------------ */
printf("\n1. Solution without using coupled local minimizer.\n\n");
/* Set various options to non-default values if required. */
nag_glopt_opt_set("Distance Tolerance = 1.0e-5", iopts, liopts, opts, lopts,
&fail);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set("Constraint Tolerance = 1.0e-4", iopts, liopts, opts,
lopts, &fail);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set("Constraint Norm = Euclidean", iopts, liopts, opts,
lopts, &fail);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set("Repeatability = On", iopts, liopts, opts, lopts,
&fail);
sprintf(optstr, "Target Objective Value = %32.16e", f_target);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set(optstr, iopts, liopts, opts, lopts, &fail);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set("Target Objective Tolerance = 1.0e-4", iopts, liopts,
opts, lopts, &fail);
if (fail.code != NE_NOERROR) {
printf("Error from nag_glopt_opt_set (e05zkc).\n%s\n", fail.message);
exit_status = 1;
goto END;
}
/* nag_glopt_nlp_pso (e05sbc).
* Global optimization using particle swarm algorithm (PSO), comprehensive.
*/
nag_glopt_nlp_pso(ndim, ncon, npar, xb, &fb, cb, bl, bu, xbest,
fbest, cbest, objfun_schwefel, confun,
monmod, iopts, opts, &comm, itt, &inform, &fail);
/* It is essential to test fail.code on exit. */
switch (fail.code) {
case NE_NOERROR:
case NW_FAST_SOLUTION:
case NW_SOLUTION_NOT_GUARANTEED:
/* No errors, best found solution at xb returned in fb. */
display_result(ndim, ncon, xb, fb, cb, itt, inform);
break;
case NE_USER_STOP:
/* Exit flag set in objfun, confun or monmod and returned in inform. */
display_result(ndim, ncon, xb, fb, cb, itt, inform);
break;
default: /* An error was detected. */
exit_status = 1;
printf("Error from nag_glopt_nlp_pso (e05sbc)\n%s\n", fail.message);
goto END;
}
/* ------------------------------------------------------------------ */
printf("2. Solution using coupled local minimizer nag_opt_nlp (e04ucc).\n\n");
/* Set the local minimizer to be nag_opt_nlp (e04ucc) and set corresponding
* options.
*/
nag_glopt_opt_set("Local Minimizer = e04ucc", iopts, liopts, opts, lopts,
&fail);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set("Local Interior Major Iterations = 15", iopts, liopts,
opts, lopts, &fail);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set("Local Interior Minor Iterations = 5", iopts, liopts,
opts, lopts, &fail);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set("Local Exterior Major Iterations = 50", iopts, liopts,
opts, lopts, &fail);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set("Local Exterior Minor Iterations = 15", iopts, liopts,
opts, lopts, &fail);
if (fail.code != NE_NOERROR) {
printf("Error from nag_glopt_opt_set (e05zkc).\n%s\n", fail.message);
exit_status = 1;
goto END;
}
/* Query the option Distance Tolerance */
nag_glopt_opt_get("Distance Tolerance", &ivalue, &rvalue, cvalue, lcvalue,
&optype, iopts, opts, &fail);
if (fail.code != NE_NOERROR) {
printf("Error from nag_glopt_opt_get (e05zlc).\n%s\n", fail.message);
exit_status = 1;
goto END;
}
/* Adjust Distance Tolerance dependent upon its current value */
rvalue = rvalue * 10.0;
sprintf(optstr, "Distance Tolerance = %32.16e", rvalue);
nag_glopt_opt_set(optstr, iopts, liopts, opts, lopts, &fail);
rvalue = 0.1 * rvalue;
sprintf(optstr, "Local Interior Tolerance = %32.16e", rvalue);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set(optstr, iopts, liopts, opts, lopts, &fail);
rvalue = rvalue * 1.0e-4;
sprintf(optstr, "Local Exterior Tolerance = %32.16e", rvalue);
if (fail.code == NE_NOERROR)
nag_glopt_opt_set(optstr, iopts, liopts, opts, lopts, &fail);
if (fail.code != NE_NOERROR) {
printf("Error from nag_glopt_opt_set (e05zkc).\n%s\n", fail.message);
exit_status = 1;
goto END;
}
/* nag_glopt_nlp_pso (e05sbc).
* Global optimization using particle swarm algorithm (PSO), comprehensive.
*/
nag_glopt_nlp_pso(ndim, ncon, npar, xb, &fb, cb, bl, bu, xbest,
fbest, cbest, objfun_schwefel, confun,
monmod, iopts, opts, &comm, itt, &inform, &fail);
/* It is essential to test fail.code on exit. */
switch (fail.code) {
case NE_NOERROR:
case NW_FAST_SOLUTION:
case NW_SOLUTION_NOT_GUARANTEED:
case NW_NOT_FEASIBLE:
/* nag_glopt_nlp_pso (e05sbc) encountered no errors during
* operation, and will have returned the best solution found.
*/
display_result(ndim, ncon, xb, fb, cb, itt, inform);
break;
case NE_USER_STOP:
/* Exit flag set in objfun, confun or monmod and returned in inform. */
display_result(ndim, ncon, xb, fb, cb, itt, inform);
break;
default: /* An error was detected. */
exit_status = 1;
printf("Error from nag_glopt_nlp_pso (e05sbc)\n%s\n", fail.message);
goto END;
}
END:
/* Clean up memory. */
NAG_FREE(bl);
NAG_FREE(bu);
NAG_FREE(cb);
NAG_FREE(cbest);
NAG_FREE(fbest);
NAG_FREE(xb);
NAG_FREE(xbest);
return exit_status;
}
static void NAG_CALL objfun_schwefel(Integer *mode, Integer ndim,
const double x[], double *objf,
double vecout[], Integer nstate,
Nag_Comm *comm)
{
/* Objective function routine returning the schwefel function and
* its gradient.
*/
Nag_Boolean evalobjf, evalobjg;
Integer i;
if (comm->user[0] == -1.0) {
printf("(User-supplied callback objfun_schwefel, first invocation.)\n");
comm->user[0] = 0.0;
}
/* Test nstate to indicate what stage of computation has been reached. */
switch (nstate) {
case 2:
/* objfun is called for the very first time. */
break;
case 1:
/* objfun is called on entry to a NAG local minimizer. */
break;
default:
/* This will be the normal value of nstate. */
;
}
/* Test mode to determine whether to calculate objf and/or objgrd. */
evalobjf = Nag_FALSE;
evalobjg = Nag_FALSE;
switch (*mode) {
case 0:
case 5:
/* Only the value of the objective function is needed. */
evalobjf = Nag_TRUE;
break;
case 1:
case 6:
/* Only the values of the ndim gradients are required. */
evalobjg = Nag_TRUE;
break;
case 2:
case 7:
/* Both the objective function and the ndim gradients are required. */
evalobjf = Nag_TRUE;
evalobjg = Nag_TRUE;
}
if (evalobjf) {
/* Evaluate the objective function. */
*objf = 0.0;
for (i = 0; i < ndim; i++)
*objf += x[i] * sin(sqrt(fabs(x[i])));
}
if (evalobjg) {
/* Calculate the gradient of the objective function, */
/* and return the result in vecout. */
for (i = 0; i < ndim; i++) {
vecout[i] = sqrt(fabs(x[i]));
vecout[i] = sin(vecout[i]) + 0.5 * vecout[i] * cos(vecout[i]);
}
}
}
static void NAG_CALL confun(Integer *mode, Integer ncon, Integer ndim,
Integer tdcj, const Integer needc[],
const double x[], double c[], double cjac[],
Integer nstate, Nag_Comm *comm)
{
/* Supplies constraints.
* cjac[(k-1)*tdcj + (i-1)] corresponds to dc[k]/dx[i]
* for k=1,...,ncon and i=1,...,ndim.
*/
Integer k;
Nag_Boolean evalc, evalcjac;
if (comm->user[1] == -1.0) {
printf("(User-supplied callback confun, first invocation.)\n");
comm->user[1] = 0.0;
}
/* Test nstate to determine whether the local minimizer is being called
* for the first time from a new start point.
*/
if (nstate == 1) {
/* Set any constant elements of the Jacobian matrix. */
cjac[0] = 3.0;
cjac[1] = -2.0;
}
/* mode: are constraints, derivatives, or both are required? */
evalc = (*mode == 0 || *mode == 2) ? Nag_TRUE : Nag_FALSE;
evalcjac = (*mode == 1 || *mode == 2) ? Nag_TRUE : Nag_FALSE;
for (k = 0; k < ncon; k++) {
if (needc[k] <= 0)
continue;
if (evalc == Nag_TRUE) {
/* Constraint values are required.
* Only those for which needc is nonzero need be set.
*/
switch (k) {
case 0:
c[k] = 3.0 * x[0] - 2.0 * x[1];
break;
case 1:
c[k] = x[0] * x[0] - x[1] * x[1] + 3.0 * x[0] * x[1];
break;
case 2:
c[k] = cos(pow((x[0] / 200.0), 2) + (x[1] / 100.0));
break;
default:
/* This constraint is not coded (there are only three).
* Terminate.
*/
*mode = -1;
break;
}
}
if (*mode < 0)
break;
if (evalcjac == Nag_TRUE) {
/* Constraint derivatives (cjac) are required. */
switch (k) {
case 0:
/* Constant derivatives set when nstate=1 remain throughout
* the local minimization.
*/
break;
case 1:
/* If the constraint derivatives are known and are readily
* calculated, populate cjac when required.
*/
cjac[k * tdcj] = 2.0 * x[0] + 3.0 * x[1];
cjac[k * tdcj + 1] = -2.0 * x[1] + 3.0 * x[0];
break;
default:
/* Any elements of cjac left unaltered will be approximated
* using finite differences when required.
*/
;
}
}
}
}
static void NAG_CALL monmod(Integer ndim, Integer ncon, Integer npar,
double x[], const double xb[], double fb,
const double cb[], const double xbest[],
const double fbest[], const double cbest[],
const Integer itt[], Nag_Comm *comm,
Integer *inform)
{
Integer i, j;
#define X(J, I) x[(J-1)*ndim + (I-1)]
#define XBEST(J, I) xbest[(J-1)*ndim + (I-1)]
#define CBEST(J, I) cbest[(J-1)*ncon + (I-1)]
if (comm->user[2] == -1.0) {
printf("(User-supplied callback monmod, first invocation.)\n");
comm->user[2] = 0.0;
}
if (detail_level) {
/* Report on the first iteration, and every report_freq iterations. */
if (itt[0] == 1 || itt[0] % report_freq == 0) {
printf("* Locations of particles\n");
for (j = 1; j <= npar; j++) {
printf(" * Particle %2" NAG_IFMT "\n", j);
for (i = 1; i <= ndim; i++)
printf(" %2" NAG_IFMT " %13.5f\n", i, X(j, i));
}
printf("* Cognitive memory\n");
for (j = 1; j <= npar; j++) {
printf(" * Particle %2" NAG_IFMT "\n", j);
printf(" * Best position\n");
for (i = 1; i <= ndim; i++)
printf(" %2" NAG_IFMT " %13.5f\n", i, XBEST(j, i));
printf(" * Function value at best position\n");
printf(" %13.5f\n", fbest[j - 1]);
printf(" * Best constraint violations\n");
for (i = 1; i <= ncon; i++)
printf(" %2" NAG_IFMT " %13.5f\n", i, CBEST(j, i));
}
printf("* Current global optimum candidate\n");
for (i = 1; i <= ndim; i++)
printf(" %2" NAG_IFMT " %13.5f\n", i, xb[i - 1]);
printf("* Current global optimum value\n");
printf(" %13.5f\n\n", fb);
printf("* Constraint violations of candidate\n");
for (i = 1; i <= ncon; i++)
printf(" %2" NAG_IFMT " %13.5f\n", i, cb[i - 1]);
}
}
/* If required set *inform<0 to force exit. */
*inform = 0;
#undef CBEST
#undef XBEST
#undef X
}
static Integer display_option(const char *optstr, const Integer iopts[],
const double opts[])
{
/* Subroutine to query optype and print the appropriate option values. */
/* Scalars */
Integer exit_status = 0, lcvalue = 17;
double rvalue = 0.0;
Integer ivalue = 0;
/* Arrays */
char cvalue[17];
/* Nag Types */
Nag_VariableType optype;
NagError fail;
INIT_FAIL(fail);
/* nag_glopt_opt_get (e05zlc).
* Option getting routine for nag_glopt_nlp_pso (e05sbc).
*/
nag_glopt_opt_get(optstr, &ivalue, &rvalue, cvalue, lcvalue, &optype, iopts,
opts, &fail);
if (fail.code != NE_NOERROR) {
printf("Error from nag_glopt_opt_get (e05zlc).\n%s\n", fail.message);
exit_status = 1;
goto END;
}
switch (optype) {
case Nag_Integer:
printf("%-38s: %13" NAG_IFMT "\n", optstr, ivalue);
break;
case Nag_Real:
printf("%-38s: %13.4f\n", optstr, rvalue);
break;
case Nag_Character:
printf("%-38s: %13s\n", optstr, cvalue);
break;
case Nag_Integer_Additional:
printf("%-38s: %13" NAG_IFMT " %16s\n", optstr, ivalue, cvalue);
break;
case Nag_Real_Additional:
printf("%-38s: %13.4f %16s\n", optstr, rvalue, cvalue);
break;
default:;
}
END:
return exit_status;
}
static void display_result(Integer ndim, Integer ncon, const double xb[],
double fb, const double cb[], const Integer itt[],
Integer inform)
{
/* Display final results in comparison to known global optimum. */
Integer i;
/* Display final counters. */
printf(" Algorithm Statistics\n");
printf(" --------------------\n");
printf("%-38s: %13" NAG_IFMT "\n", "Total complete iterations", itt[0]);
printf("%-38s: %13" NAG_IFMT "\n", "Complete iterations since improvement",
itt[1]);
printf("%-38s: %13" NAG_IFMT "\n", "Total particles converged to xb",
itt[2]);
printf("%-38s: %13" NAG_IFMT "\n", "Total improvements to global optimum",
itt[3]);
printf("%-38s: %13" NAG_IFMT "\n", "Total function evaluations", itt[4]);
printf("%-38s: %13" NAG_IFMT "\n", "Total particles re-initialized",
itt[5]);
printf("%-38s: %13" NAG_IFMT "\n\n", "Total constraints violated", itt[6]);
/* Display why finalization occurred. */
switch (inform) {
case 1:
printf("Solution Status : Target value achieved\n");
break;
case 2:
printf("Solution Status : Minimum swarm standard deviation obtained\n");
break;
case 3:
printf("Solution Status : Sufficient number of particles converged\n");
break;
case 4:
printf("Solution Status : Maximum static iterations attained\n");
break;
case 5:
printf("Solution Status : Maximum complete iterations attained\n");
break;
case 6:
printf("Solution Status : Maximum function evaluations exceeded\n");
break;
case 7:
printf("Solution Status : Feasible point located\n");
break;
default:
if (inform < 0) {
printf("Solution Status : User termination, inform = %16" NAG_IFMT "\n",
inform);
return;
}
printf("Solution Status : Termination, an error has been detected\n");
break;
}
/* Display final objective value and location. */
printf(" Known constrained objective optimum : %13.3f\n", f_target);
printf(" Achieved objective value : %13.3f\n\n", fb);
printf(" Comparison between the known optimum and the achieved solution.\n");
printf(" x_target xb\n");
for (i = 0; i < ndim; i++)
printf(" %2" NAG_IFMT " %12.2f %12.2f\n", i + 1, x_target[i], xb[i]);
printf("\n");
if (ncon > 0) {
printf(" Comparison between scaled constraint violations.\n");
printf(" c_target cb\n");
for (i = 0; i < ncon; i++)
printf(" %2" NAG_IFMT " %12.5f %12.5f\n", i + 1,
c_target[i] / c_scale[i], cb[i] / c_scale[i]);
printf("\n");
}
}