cgal/QP_solver/test/QP_solver/test_random_qp2.cpp

159 lines
4.2 KiB
C++

#include <cstdlib>
#include <cassert>
#include <iostream>
#include <CGAL/Random.h>
#include <CGAL/QP_models.h>
#include <CGAL/QP_options.h>
#include <CGAL/QP_functions.h>
// choose exact floating point type
#ifndef CGAL_USE_GMP
#include <CGAL/MP_Float.h>
typedef CGAL::MP_Float ET;
#else
#include <CGAL/Gmpzf.h>
typedef CGAL::Gmpzf ET;
#endif
// program and solution types
typedef CGAL::Quadratic_program<double> Program;
typedef CGAL::Quadratic_program_solution<ET> Solution;
// random number generator
CGAL::Random rd;
CGAL::Comparison_result random_rel()
{
int z = rd.get_int(-1,2);
return CGAL::Comparison_result(z);
}
void statistics (const Solution& s,
unsigned int& o, unsigned int& i, unsigned int& u)
{
switch (s.status()) {
case CGAL::QP_OPTIMAL:
o++;
break;
case CGAL::QP_INFEASIBLE:
i++;
break;
case CGAL::QP_UNBOUNDED:
u++;
break;
default:
assert(false);
}
}
unsigned int qp_optimal = 0;
unsigned int qp_infeasible = 0;
unsigned int qp_unbounded = 0;
unsigned int nqp_optimal = 0;
unsigned int nqp_infeasible = 0;
unsigned int nqp_unbounded = 0;
unsigned int lp_optimal = 0;
unsigned int lp_infeasible = 0;
unsigned int lp_unbounded = 0;
unsigned int nlp_optimal = 0;
unsigned int nlp_infeasible = 0;
unsigned int nlp_unbounded = 0;
// parameters
int tries = 5000;
int max_dim = 11; // must be >1
int max_entry = 11; // must be >0
int main() {
// print seed
std::cout << "Random seed: " << rd.get_seed() << std::endl;
// options
CGAL::Quadratic_program_options options;
options.set_auto_validation(true);
// generate a set of small random qp's
for (int i=0; i<tries; ++i) {
// first choose dimensions
int n = rd.get_int(1,max_dim);
int m = rd.get_int(1,max_dim);
// construct matrix D as C^T C, for C randomly chosen with n columns
int k = rd.get_int (1, 2*n); // number of rows of C
std::vector<std::vector<int> > C (k, std::vector<int>(n, 0));
for (int j=0; j<k+n; ++j) // sparse C
C[rd.get_int(0, k)][rd.get_int(0,n)] =
rd.get_int(-max_entry, max_entry);
// now fill the program
Program p;
// A
for (int j=0; j<n+m; ++j)
p.set_a (rd.get_int(0,n), rd.get_int(0,m), rd.get_double());
// b, r
for (int i=0; i<m/2; ++i) {
p.set_b (rd.get_int(0,m), rd.get_double());
p.set_r (rd.get_int(0,m), random_rel());
}
// fl, l, fu, u
for (int j=0; j<n; ++j) {
double l = rd.get_double();
double u = rd.get_double();
if (l > u) std::swap (l, u);
p.set_l(j, rd.get_bool(), l);
p.set_u(j, rd.get_bool(), u);
}
// D
for (int i=0; i<n; ++i)
for (int j=0; j<=i; ++j) {
double entry = 0;
for (int l=0; l<k; ++l)
entry += C[l][i] * C[l][j];
p.set_d(i, j, entry);
}
// c
for (int j=0; j<n/2; ++j)
p.set_c (rd.get_int(0, n), rd.get_double());
// c0
p.set_c0(rd.get_double());
// solve it
Solution s = CGAL::solve_quadratic_program (p, ET(), options);
assert(s.is_valid());
statistics (s, qp_optimal, qp_infeasible, qp_unbounded);
// also solve it as nqp, lp, nlp
s = CGAL::solve_nonnegative_quadratic_program (p, ET(), options);
assert(s.is_valid());
statistics (s, nqp_optimal, nqp_infeasible, nqp_unbounded);
s = CGAL::solve_linear_program (p, ET(), options);
assert(s.is_valid());
statistics (s, lp_optimal, lp_infeasible, lp_unbounded);
s = CGAL::solve_nonnegative_linear_program (p, ET(), options);
assert(s.is_valid());
statistics (s, nlp_optimal, nlp_infeasible, nlp_unbounded);
}
// output statistics
std::cout << "Solved " << tries
<< " random QP / NQP / LP / NLP .\n"
<< " Optimal: "
<< qp_optimal << " / "
<< nqp_optimal << " / "
<< lp_optimal << " / "
<< nlp_optimal << "\n"
<< " Infeasible: "
<< qp_infeasible << " / "
<< nqp_infeasible << " / "
<< lp_infeasible << " / "
<< nlp_infeasible << "\n"
<< " Unbounded: "
<< qp_unbounded << " / "
<< nqp_unbounded << " / "
<< lp_unbounded << " / "
<< nlp_unbounded << std::endl;
return 0;
}