int main(void) {
glp_prob *lp;
int ia[1+1000], ja[1+1000];
double ar[1+1000], z, x1, x2, x3;
s1: lp = glp_create_prob();
s2: glp_set_prob_name(lp, "sample");
s3: glp_set_obj_dir(lp, GLP_MAX);
s4: glp_add_rows(lp, 3);
s5: glp_set_row_name(lp, 1, "p");
s6: glp_set_row_bnds(lp, 1, GLP_UP, 0.0, 100.0);
s7: glp_set_row_name(lp, 2, "q");
s8: glp_set_row_bnds(lp, 2, GLP_UP, 0.0, 600.0);
s9: glp_set_row_name(lp, 3, "r");
s10: glp_set_row_bnds(lp, 3, GLP_UP, 0.0, 300.0);
s11: glp_add_cols(lp, 3);
s12: glp_set_col_name(lp, 1, "x1");
s13: glp_set_col_bnds(lp, 1, GLP_LO, 0.0, 0.0);
s14: glp_set_obj_coef(lp, 1, 10.0);
s15: glp_set_col_name(lp, 2, "x2");
s16: glp_set_col_bnds(lp, 2, GLP_LO, 0.0, 0.0);
s17: glp_set_obj_coef(lp, 2, 6.0);
s18: glp_set_col_name(lp, 3, "x3");
s19: glp_set_col_bnds(lp, 3, GLP_LO, 0.0, 0.0);
s20: glp_set_obj_coef(lp, 3, 4.0);
s21: ia[1] = 1, ja[1] = 1, ar[1] = 1.0; /* a[1,1] = 1 */
s22: ia[2] = 1, ja[2] = 2, ar[2] = 1.0; /* a[1,2] = 1 */
s23: ia[3] = 1, ja[3] = 3, ar[3] = 1.0; /* a[1,3] = 1 */
s24: ia[4] = 2, ja[4] = 1, ar[4] = 10.0; /* a[2,1] = 10 */
s25: ia[5] = 3, ja[5] = 1, ar[5] = 2.0; /* a[3,1] = 2 */
s26: ia[6] = 2, ja[6] = 2, ar[6] = 4.0; /* a[2,2] = 4 */
s27: ia[7] = 3, ja[7] = 2, ar[7] = 2.0; /* a[3,2] = 2 */
s28: ia[8] = 2, ja[8] = 3, ar[8] = 5.0; /* a[2,3] = 5 */
s29: ia[9] = 3, ja[9] = 3, ar[9] = 6.0; /* a[3,3] = 6 */
s30: glp_load_matrix(lp, 9, ia, ja, ar);
s31: glp_simplex(lp, NULL);
s32: z = glp_get_obj_val(lp);
s33: x1 = glp_get_col_prim(lp, 1);
s34: x2 = glp_get_col_prim(lp, 2);
s35: x3 = glp_get_col_prim(lp, 3);
s36: printf("\nz = %g; x1 = %g; x2 = %g; x3 = %g\n", z, x1, x2, x3);
s37: glp_delete_prob(lp);
return 0;
}
double CMyProblem::RealResult()
{
char buff[256];
int i=1;
int j=1;
double value;
int add;
int result=0;
int col_num=glp_find_col(lp,GenerateColName(buff, "x", i, j));
while( col_num )
{
value = glp_get_col_prim(lp,col_num);
if ( abs(1.0-value)<AVAIL_ERROR )
{
add=j*w[i-1];
}
j++;
col_num=glp_find_col(lp,GenerateColName(buff, "x", i, j));
if (!col_num)
{
result+=add;
i++;
j=1;
col_num=glp_find_col(lp,GenerateColName(buff, "x", i, j));
}
}
return result;
}
void printResult(glp_prob * Prob, FILE * out) {
int i;
char buf[AUXSIZE];
glp_smcp * param = malloc(sizeof(glp_smcp));
glp_init_smcp(param);
param->msg_lev = GLP_MSG_ERR;
param->presolve = GLP_ON;
int solution[MAXSIZE];
for( i = 0 ; i < n ; ++i )
solution[planes[i].pos] = i;
mapSolution(Prob,solution);
glp_simplex(Prob,param);
if( simpleOutput ) {
time(&final);
fprintf(out,"%i & %i \\\\ \n",(int)glp_get_obj_val(Prob),(int)difftime(final,initial));
return;
}
fprintf(out,"Best found solution's value: %lf\n\n",glp_get_obj_val(Prob));
double time;
for( i = 0 ; i < n ; ++i ) {
sprintf(buf,"x%i",solution[i]);
time = glp_get_col_prim(Prob, glp_find_col(Prob, buf));
fprintf(out,"The %i-th airplane to arrive is airplane %i, at the time %lf\n",
i+1,solution[i]+1,time);
}
}
void glpk_wrapper::get_solution(box & b) {
assert(is_sat());
for (unsigned int i = 0; i < b.size(); i++) {
if (solver_type == SIMPLEX || solver_type == EXACT) {
b[i] = glp_get_col_prim(lp, i+1);
} else {
assert(solver_type == INTERIOR);
b[i] = glp_ipt_col_prim(lp, i+1);
}
}
}
static int branch_first(glp_tree *T, int *_next)
{ int j, next;
double beta;
/* choose the column to branch on */
for (j = 1; j <= T->n; j++)
if (T->non_int[j]) break;
xassert(1 <= j && j <= T->n);
/* select the branch to be solved next */
beta = glp_get_col_prim(T->mip, j);
if (beta - floor(beta) < ceil(beta) - beta)
next = GLP_DN_BRNCH;
else
next = GLP_UP_BRNCH;
*_next = next;
return j;
}
static int branch_mostf(glp_tree *T, int *_next)
{ int j, jj, next;
double beta, most, temp;
/* choose the column to branch on */
jj = 0, most = DBL_MAX;
for (j = 1; j <= T->n; j++)
{ if (T->non_int[j])
{ beta = glp_get_col_prim(T->mip, j);
temp = floor(beta) + 0.5;
if (most > fabs(beta - temp))
{ jj = j, most = fabs(beta - temp);
if (beta < temp)
next = GLP_DN_BRNCH;
else
next = GLP_UP_BRNCH;
}
}
}
*_next = next;
return jj;
}
inline void PrintSolArray(glp_prob* lp, char* name, ostream &out, bool integer=false)
{
out << name << endl;
char buff[256];
int i=1;
int j=1;
double value;
int col_num=glp_find_col(lp,GenerateColName(buff, name, i, j));
while( col_num )
{
if (!integer)
{
value = glp_get_col_prim(lp,col_num);
//проверка иксов на целочисленность
if ((strcmp(name,"x")==0)&&(IsNotBinary(value)))
{
out<<"!!! ";
}
}
else
{
value = glp_mip_col_val(lp,col_num);
}
out << value << ";";
//допишем коэффициенты ещё
//out << "*(" << glp_get_obj_coef(lp,col_num) << ");";
j++;
col_num=glp_find_col(lp,GenerateColName(buff, name, i, j));
if (!col_num)
{
out << endl;
i++;
j=1;
col_num=glp_find_col(lp,GenerateColName(buff, name, i, j));
}
}
}
vector<int> LinearProblem::ElasticFilter() const {
LinearProblem tmp(*this);
int realCols = glp_get_num_cols(tmp.lp_);
for (int i = realCols; i > 0; --i) { // set old coefs to zero
glp_set_obj_coef(tmp.lp_, i, 0);
}
int elasticCols = glp_get_num_rows(tmp.lp_);
glp_add_cols(tmp.lp_, elasticCols);
for (int i = 1; i <= elasticCols; ++i) {
int indices[MAX_VARS];
double values[MAX_VARS];
int nonZeros = glp_get_mat_row(tmp.lp_, i, indices, values);
indices[nonZeros + 1] = realCols + i;
values[nonZeros + 1] = 1.0;
glp_set_mat_row(tmp.lp_, i, nonZeros + 1, indices, values);
glp_set_obj_coef(tmp.lp_, realCols + i, 1);
glp_set_col_bnds(tmp.lp_, realCols + i, GLP_UP, 0.0, 0.0);
}
vector<int> suspects;
glp_std_basis(tmp.lp_);
int status = tmp.Solve();
while ((status != GLP_INFEAS) && (status != GLP_NOFEAS)) {
for (int i = 1; i <= elasticCols; ++i) {
if (glp_get_col_prim(tmp.lp_, realCols + i) < 0) {
suspects.push_back(i);
glp_set_col_bnds(tmp.lp_, realCols + i, GLP_FX, 0.0, 0.0);
}
}
status = tmp.Solve();
}
return suspects;
}
inline void GetSolArray(glp_prob* lp, char* name, vector<vector<double>> &arr, bool integer=false)
{
arr.clear();
char buff[256];
int i=1;
int j=1;
double value;
glp_create_index(lp);
int col_num=glp_find_col(lp,GenerateColName(buff, name, i, j));
vector<double> row;
while( col_num )
{
if (!integer)
{
value = glp_get_col_prim(lp,col_num);
}
else
{
value = glp_mip_col_val(lp,col_num);
}
row.push_back(value);
//cout << value << ";";
j++;
col_num=glp_find_col(lp,GenerateColName(buff, name, i, j));
if (!col_num)
{
arr.push_back(row);
//cout << endl;
row.clear();
i++;
j=1;
col_num=glp_find_col(lp,GenerateColName(buff, name, i, j));
}
}
}
int glp_mpl_postsolve(glp_tran *tran, glp_prob *prob, int sol)
{ /* postsolve the model */
int j, m, n, ret;
double x;
if (!(tran->phase == 3 && !tran->flag_p))
xerror("glp_mpl_postsolve: invalid call sequence\n");
if (!(sol == GLP_SOL || sol == GLP_IPT || sol == GLP_MIP))
xerror("glp_mpl_postsolve: sol = %d; invalid parameter\n",
sol);
m = mpl_get_num_rows(tran);
n = mpl_get_num_cols(tran);
if (!(m == glp_get_num_rows(prob) &&
n == glp_get_num_cols(prob)))
xerror("glp_mpl_postsolve: wrong problem object\n");
if (!mpl_has_solve_stmt(tran))
{ ret = 0;
goto done;
}
for (j = 1; j <= n; j++)
{ if (sol == GLP_SOL)
x = glp_get_col_prim(prob, j);
else if (sol == GLP_IPT)
x = glp_ipt_col_prim(prob, j);
else if (sol == GLP_MIP)
x = glp_mip_col_val(prob, j);
else
xassert(sol != sol);
if (fabs(x) < 1e-9) x = 0.0;
mpl_put_col_value(tran, j, x);
}
ret = mpl_postsolve(tran);
if (ret == 3)
ret = 0;
else if (ret == 4)
ret = 1;
done: return ret;
}
OptSolutionData* GLPKRunSolver(int ProbType) {
OptSolutionData* NewSolution = NULL;
int NumVariables = glp_get_num_cols(GLPKModel);
int Status = 0;
if (ProbType == MILP) {
Status = glp_simplex(GLPKModel, NULL); // Use default settings
if (Status != 0) {
FErrorFile() << "Failed to optimize problem." << endl;
FlushErrorFile();
return NULL;
}
Status = glp_intopt(GLPKModel, NULL); // Use default settings
if (Status != 0) {
FErrorFile() << "Failed to optimize problem." << endl;
FlushErrorFile();
return NULL;
}
NewSolution = new OptSolutionData;
Status = glp_mip_status(GLPKModel);
if (Status == GLP_UNDEF || Status == GLP_NOFEAS) {
NewSolution->Status = INFEASIBLE;
return NewSolution;
} else if (Status == GLP_FEAS) {
NewSolution->Status = UNBOUNDED;
return NewSolution;
} else if (Status == GLP_OPT) {
NewSolution->Status = SUCCESS;
} else {
delete NewSolution;
FErrorFile() << "Problem status unrecognized." << endl;
FlushErrorFile();
return NULL;
}
NewSolution->Objective = glp_mip_obj_val(GLPKModel);
NewSolution->SolutionData.resize(NumVariables);
for (int i=0; i < NumVariables; i++) {
NewSolution->SolutionData[i] = glp_mip_col_val(GLPKModel, i+1);
}
} else if (ProbType == LP) {
//First we check the basis matrix to ensure it is not singular
if (glp_warm_up(GLPKModel) != 0) {
glp_adv_basis(GLPKModel, 0);
}
Status = glp_simplex(GLPKModel, NULL); // Use default settings
if (Status == GLP_EBADB) { /* the basis is invalid; build some valid basis */
glp_adv_basis(GLPKModel, 0);
Status = glp_simplex(GLPKModel, NULL); // Use default settings
}
if (Status != 0) {
FErrorFile() << "Failed to optimize problem." << endl;
FlushErrorFile();
return NULL;
}
NewSolution = new OptSolutionData;
Status = glp_get_status(GLPKModel);
if (Status == GLP_INFEAS || Status == GLP_NOFEAS || Status == GLP_UNDEF) {
cout << "Model is infeasible" << endl;
FErrorFile() << "Model is infeasible" << endl;
FlushErrorFile();
NewSolution->Status = INFEASIBLE;
return NewSolution;
} else if (Status == GLP_FEAS || Status == GLP_UNBND) {
cout << "Model is unbounded" << endl;
FErrorFile() << "Model is unbounded" << endl;
FlushErrorFile();
NewSolution->Status = UNBOUNDED;
return NewSolution;
} else if (Status == GLP_OPT) {
NewSolution->Status = SUCCESS;
} else {
delete NewSolution;
FErrorFile() << "Problem status unrecognized." << endl;
FlushErrorFile();
return NULL;
}
NewSolution->Objective = glp_get_obj_val(GLPKModel);
NewSolution->SolutionData.resize(NumVariables);
for (int i=0; i < NumVariables; i++) {
NewSolution->SolutionData[i] = glp_get_col_prim(GLPKModel, i+1);
}
} else {
FErrorFile() << "Optimization problem type cannot be handled by GLPK solver." << endl;
FlushErrorFile();
return NULL;
}
return NewSolution;
}
double lpx_get_col_prim(LPX *lp, int j)
{ /* retrieve column primal value (basic solution) */
return glp_get_col_prim(lp, j);
}
static int branch_drtom(glp_tree *T, int *_next)
{ glp_prob *mip = T->mip;
int m = mip->m;
int n = mip->n;
char *non_int = T->non_int;
int j, jj, k, t, next, kase, len, stat, *ind;
double x, dk, alfa, delta_j, delta_k, delta_z, dz_dn, dz_up,
dd_dn, dd_up, degrad, *val;
/* basic solution of LP relaxation must be optimal */
xassert(glp_get_status(mip) == GLP_OPT);
/* allocate working arrays */
ind = xcalloc(1+n, sizeof(int));
val = xcalloc(1+n, sizeof(double));
/* nothing has been chosen so far */
jj = 0, degrad = -1.0;
/* walk through the list of columns (structural variables) */
for (j = 1; j <= n; j++)
{ /* if j-th column is not marked as fractional, skip it */
if (!non_int[j]) continue;
/* obtain (fractional) value of j-th column in basic solution
of LP relaxation */
x = glp_get_col_prim(mip, j);
/* since the value of j-th column is fractional, the column is
basic; compute corresponding row of the simplex table */
len = glp_eval_tab_row(mip, m+j, ind, val);
/* the following fragment computes a change in the objective
function: delta Z = new Z - old Z, where old Z is the
objective value in the current optimal basis, and new Z is
the objective value in the adjacent basis, for two cases:
1) if new upper bound ub' = floor(x[j]) is introduced for
j-th column (down branch);
2) if new lower bound lb' = ceil(x[j]) is introduced for
j-th column (up branch);
since in both cases the solution remaining dual feasible
becomes primal infeasible, one implicit simplex iteration
is performed to determine the change delta Z;
it is obvious that new Z, which is never better than old Z,
is a lower (minimization) or upper (maximization) bound of
the objective function for down- and up-branches. */
for (kase = -1; kase <= +1; kase += 2)
{ /* if kase < 0, the new upper bound of x[j] is introduced;
in this case x[j] should decrease in order to leave the
basis and go to its new upper bound */
/* if kase > 0, the new lower bound of x[j] is introduced;
in this case x[j] should increase in order to leave the
basis and go to its new lower bound */
/* apply the dual ratio test in order to determine which
auxiliary or structural variable should enter the basis
to keep dual feasibility */
k = glp_dual_rtest(mip, len, ind, val, kase, 1e-9);
if (k != 0) k = ind[k];
/* if no non-basic variable has been chosen, LP relaxation
of corresponding branch being primal infeasible and dual
unbounded has no primal feasible solution; in this case
the change delta Z is formally set to infinity */
if (k == 0)
{ delta_z =
(T->mip->dir == GLP_MIN ? +DBL_MAX : -DBL_MAX);
goto skip;
}
/* row of the simplex table that corresponds to non-basic
variable x[k] choosen by the dual ratio test is:
x[j] = ... + alfa * x[k] + ...
where alfa is the influence coefficient (an element of
the simplex table row) */
/* determine the coefficient alfa */
for (t = 1; t <= len; t++) if (ind[t] == k) break;
xassert(1 <= t && t <= len);
alfa = val[t];
/* since in the adjacent basis the variable x[j] becomes
non-basic, knowing its value in the current basis we can
determine its change delta x[j] = new x[j] - old x[j] */
delta_j = (kase < 0 ? floor(x) : ceil(x)) - x;
/* and knowing the coefficient alfa we can determine the
corresponding change delta x[k] = new x[k] - old x[k],
where old x[k] is a value of x[k] in the current basis,
and new x[k] is a value of x[k] in the adjacent basis */
delta_k = delta_j / alfa;
/* Tomlin noticed that if the variable x[k] is of integer
kind, its change cannot be less (eventually) than one in
the magnitude */
if (k > m && glp_get_col_kind(mip, k-m) != GLP_CV)
{ /* x[k] is structural integer variable */
if (fabs(delta_k - floor(delta_k + 0.5)) > 1e-3)
{ if (delta_k > 0.0)
delta_k = ceil(delta_k); /* +3.14 -> +4 */
else
delta_k = floor(delta_k); /* -3.14 -> -4 */
}
}
/* now determine the status and reduced cost of x[k] in the
current basis */
if (k <= m)
{ stat = glp_get_row_stat(mip, k);
dk = glp_get_row_dual(mip, k);
}
else
{ stat = glp_get_col_stat(mip, k-m);
dk = glp_get_col_dual(mip, k-m);
}
/* if the current basis is dual degenerate, some reduced
costs which are close to zero may have wrong sign due to
round-off errors, so correct the sign of d[k] */
switch (T->mip->dir)
{ case GLP_MIN:
if (stat == GLP_NL && dk < 0.0 ||
stat == GLP_NU && dk > 0.0 ||
stat == GLP_NF) dk = 0.0;
break;
case GLP_MAX:
if (stat == GLP_NL && dk > 0.0 ||
stat == GLP_NU && dk < 0.0 ||
stat == GLP_NF) dk = 0.0;
break;
default:
xassert(T != T);
}
/* now knowing the change of x[k] and its reduced cost d[k]
we can compute the corresponding change in the objective
function delta Z = new Z - old Z = d[k] * delta x[k];
note that due to Tomlin's modification new Z can be even
worse than in the adjacent basis */
delta_z = dk * delta_k;
skip: /* new Z is never better than old Z, therefore the change
delta Z is always non-negative (in case of minimization)
or non-positive (in case of maximization) */
switch (T->mip->dir)
{ case GLP_MIN: xassert(delta_z >= 0.0); break;
case GLP_MAX: xassert(delta_z <= 0.0); break;
default: xassert(T != T);
}
/* save the change in the objective fnction for down- and
up-branches, respectively */
if (kase < 0) dz_dn = delta_z; else dz_up = delta_z;
}
/* thus, in down-branch no integer feasible solution can be
better than Z + dz_dn, and in up-branch no integer feasible
solution can be better than Z + dz_up, where Z is value of
the objective function in the current basis */
/* following the heuristic by Driebeck and Tomlin we choose a
column (i.e. structural variable) which provides largest
degradation of the objective function in some of branches;
besides, we select the branch with smaller degradation to
be solved next and keep other branch with larger degradation
in the active list hoping to minimize the number of further
backtrackings */
if (degrad < fabs(dz_dn) || degrad < fabs(dz_up))
{ jj = j;
if (fabs(dz_dn) < fabs(dz_up))
{ /* select down branch to be solved next */
next = GLP_DN_BRNCH;
degrad = fabs(dz_up);
}
else
{ /* select up branch to be solved next */
next = GLP_UP_BRNCH;
degrad = fabs(dz_dn);
}
/* save the objective changes for printing */
dd_dn = dz_dn, dd_up = dz_up;
/* if down- or up-branch has no feasible solution, we does
not need to consider other candidates (in principle, the
corresponding branch could be pruned right now) */
if (degrad == DBL_MAX) break;
}
}
/* free working arrays */
xfree(ind);
xfree(val);
/* something must be chosen */
xassert(1 <= jj && jj <= n);
#if 1 /* 02/XI-2009 */
if (degrad < 1e-6 * (1.0 + 0.001 * fabs(mip->obj_val)))
{ jj = branch_mostf(T, &next);
goto done;
}
#endif
if (T->parm->msg_lev >= GLP_MSG_DBG)
{ xprintf("branch_drtom: column %d chosen to branch on\n", jj);
if (fabs(dd_dn) == DBL_MAX)
xprintf("branch_drtom: down-branch is infeasible\n");
else
xprintf("branch_drtom: down-branch bound is %.9e\n",
lpx_get_obj_val(mip) + dd_dn);
if (fabs(dd_up) == DBL_MAX)
xprintf("branch_drtom: up-branch is infeasible\n");
else
xprintf("branch_drtom: up-branch bound is %.9e\n",
lpx_get_obj_val(mip) + dd_up);
}
done: *_next = next;
return jj;
}
int glp_mpl_postsolve(glp_tran *tran, glp_prob *prob, int sol)
{ /* postsolve the model */
int i, j, m, n, stat, ret;
double prim, dual;
if (!(tran->phase == 3 && !tran->flag_p))
xerror("glp_mpl_postsolve: invalid call sequence\n");
if (!(sol == GLP_SOL || sol == GLP_IPT || sol == GLP_MIP))
xerror("glp_mpl_postsolve: sol = %d; invalid parameter\n",
sol);
m = mpl_get_num_rows(tran);
n = mpl_get_num_cols(tran);
if (!(m == glp_get_num_rows(prob) &&
n == glp_get_num_cols(prob)))
xerror("glp_mpl_postsolve: wrong problem object\n");
if (!mpl_has_solve_stmt(tran))
{ ret = 0;
goto done;
}
for (i = 1; i <= m; i++)
{ if (sol == GLP_SOL)
{ stat = glp_get_row_stat(prob, i);
prim = glp_get_row_prim(prob, i);
dual = glp_get_row_dual(prob, i);
}
else if (sol == GLP_IPT)
{ stat = 0;
prim = glp_ipt_row_prim(prob, i);
dual = glp_ipt_row_dual(prob, i);
}
else if (sol == GLP_MIP)
{ stat = 0;
prim = glp_mip_row_val(prob, i);
dual = 0.0;
}
else
xassert(sol != sol);
if (fabs(prim) < 1e-9) prim = 0.0;
if (fabs(dual) < 1e-9) dual = 0.0;
mpl_put_row_soln(tran, i, stat, prim, dual);
}
for (j = 1; j <= n; j++)
{ if (sol == GLP_SOL)
{ stat = glp_get_col_stat(prob, j);
prim = glp_get_col_prim(prob, j);
dual = glp_get_col_dual(prob, j);
}
else if (sol == GLP_IPT)
{ stat = 0;
prim = glp_ipt_col_prim(prob, j);
dual = glp_ipt_col_dual(prob, j);
}
else if (sol == GLP_MIP)
{ stat = 0;
prim = glp_mip_col_val(prob, j);
dual = 0.0;
}
else
xassert(sol != sol);
if (fabs(prim) < 1e-9) prim = 0.0;
if (fabs(dual) < 1e-9) dual = 0.0;
mpl_put_col_soln(tran, j, stat, prim, dual);
}
ret = mpl_postsolve(tran);
if (ret == 3)
ret = 0;
else if (ret == 4)
ret = 1;
done: return ret;
}
static bool Graph_solve(Graph& graph, size_t loops, PositionList* position_tbl) {
glp_prob* lp = glp_create_prob();
glp_set_prob_name(lp, "scaffold");
glp_set_obj_dir(lp, GLP_MIN);
size_t rows = 0, cols = 0, vals = 0;
for (Graph::const_iterator i = graph.begin(); i != graph.end(); ++i) {
++cols; // var x_i
for (Children::const_iterator j = i->second.children.begin(); j != i->second.children.end(); ++j) {
rows += 3;
cols += 2; // var e_i_j; var E_i_j;
vals += 7;
}
}
glp_add_rows(lp, rows);
glp_add_cols(lp, cols);
std::map< std::pair< size_t, size_t >, size_t > mapping;
{
size_t row = 1, col = 1;
for (Graph::const_iterator i = graph.begin(); i != graph.end(); ++i) {
mapping[std::make_pair(i->first, -1)] = col;
glp_set_col_bnds(lp, col++, GLP_LO, 0.0, 0.0); // x_i >= 0
for (Children::const_iterator j = i->second.children.begin(); j != i->second.children.end(); ++j) {
glp_set_row_bnds(lp, row++, GLP_FX, j->second, j->second);
glp_set_row_bnds(lp, row++, GLP_LO, 0.0, 0.0);
glp_set_row_bnds(lp, row++, GLP_LO, 0.0, 0.0);
mapping[std::make_pair(i->first, j->first)] = col;
++col; // var e_i
glp_set_obj_coef(lp, col++, 1.0);
}
}
}
LOG4CXX_TRACE(logger, boost::format(" rows = %d, cols = %d, vals = %d") % rows % cols % vals);
int* ia = new int[vals + 1];
int* ja = new int[vals + 1];
double* ra = new double[vals + 1];
{
size_t l = 1, row = 1, col = 1;
for (Graph::const_iterator i = graph.begin(); i != graph.end(); ++i) {
for (Children::const_iterator j = i->second.children.begin(); j != i->second.children.end(); ++j) {
// x_j - x_i + e_i_j = d_i_j
ia[l] = row;
ja[l] = mapping[std::make_pair(j->first, -1)];
ra[l] = 1.0;
++l;
ia[l] = row;
ja[l] = mapping[std::make_pair(i->first, -1)];
ra[l] = -1.0;
++l;
ia[l] = row;
ja[l] = mapping[std::make_pair(i->first, j->first)] + 0;
ra[l] = 1.0;
++l;
++row;
// E_i_j + e_i_j >= 0
ia[l] = row;
ja[l] = mapping[std::make_pair(i->first, j->first)] + 1;
ra[l] = 1.0;
++l;
ia[l] = row;
ja[l] = mapping[std::make_pair(i->first, j->first)] + 0;
ra[l] = 1.0;
++l;
++row;
// E_i_j - e_i_j >= 0
ia[l] = row;
ja[l] = mapping[std::make_pair(i->first, j->first)] + 1;
ra[l] = 1.0;
++l;
ia[l] = row;
ja[l] = mapping[std::make_pair(i->first, j->first)] + 0;
ra[l] = -1.0;
++l;
++row;
}
}
}
glp_load_matrix(lp, vals, ia, ja, ra);
glp_smcp parm;
glp_init_smcp(&parm);
parm.it_lim = loops;
//parm.pricing = GLP_PT_PSE;
//parm.presolve = GLP_ON;
parm.msg_lev = GLP_MSG_ERR;
glp_simplex(lp, &parm);
double z = glp_get_obj_val(lp);
LOG4CXX_TRACE(logger, boost::format("z = %f") % z);
for (Graph::const_iterator i = graph.begin(); i != graph.end(); ++i) {
size_t val = glp_get_col_prim(lp, mapping[std::make_pair(i->first, -1)]);
if (position_tbl != NULL) {
(*position_tbl)[i->first] = val;
}
LOG4CXX_TRACE(logger, boost::format("x\t%d\t%d") % i->first % val);
}
delete[] ra;
delete[] ja;
delete[] ia;
glp_delete_prob(lp);
return true;
}
int glpk (int sense, int n, int m, double *c, int nz, int *rn, int *cn,
double *a, double *b, char *ctype, int *freeLB, double *lb,
int *freeUB, double *ub, int *vartype, int isMIP, int lpsolver,
int save_pb, char *save_filename, char *filetype,
double *xmin, double *fmin, double *status,
double *lambda, double *redcosts, double *time, double *mem)
{
int typx = 0;
int method;
clock_t t_start = clock();
// Obsolete
//lib_set_fault_hook (NULL, glpk_fault_hook);
//Redirect standard output
if (glpIntParam[0] > 1) glp_term_hook (glpk_print_hook, NULL);
else glp_term_hook (NULL, NULL);
//-- Create an empty LP/MILP object
glp_prob *lp = glp_create_prob ();
//-- Set the sense of optimization
if (sense == 1)
glp_set_obj_dir (lp, GLP_MIN);
else
glp_set_obj_dir (lp, GLP_MAX);
//-- Define the number of unknowns and their domains.
glp_add_cols (lp, n);
for (int i = 0; i < n; i++)
{
//-- Define type of the structural variables
if (! freeLB[i] && ! freeUB[i])
glp_set_col_bnds (lp, i+1, GLP_DB, lb[i], ub[i]);
else
{
if (! freeLB[i] && freeUB[i])
glp_set_col_bnds (lp, i+1, GLP_LO, lb[i], ub[i]);
else
{
if (freeLB[i] && ! freeUB[i])
glp_set_col_bnds (lp, i+1, GLP_UP, lb[i], ub[i]);
else
glp_set_col_bnds (lp, i+1, GLP_FR, lb[i], ub[i]);
}
}
// -- Set the objective coefficient of the corresponding
// -- structural variable. No constant term is assumed.
glp_set_obj_coef(lp,i+1,c[i]);
if (isMIP)
glp_set_col_kind (lp, i+1, vartype[i]);
}
glp_add_rows (lp, m);
for (int i = 0; i < m; i++)
{
/* If the i-th row has no lower bound (types F,U), the
corrispondent parameter will be ignored.
If the i-th row has no upper bound (types F,L), the corrispondent
parameter will be ignored.
If the i-th row is of S type, the i-th LB is used, but
the i-th UB is ignored.
*/
switch (ctype[i])
{
case 'F': typx = GLP_FR; break;
// upper bound
case 'U': typx = GLP_UP; break;
// lower bound
case 'L': typx = GLP_LO; break;
// fixed constraint
case 'S': typx = GLP_FX; break;
// double-bounded variable
case 'D': typx = GLP_DB; break;
}
glp_set_row_bnds (lp, i+1, typx, b[i], b[i]);
}
// Load constraint matrix A
glp_load_matrix (lp, nz, rn, cn, a);
// Save problem
if (save_pb) {
if (!strcmp(filetype,"cplex")){
if (lpx_write_cpxlp (lp, save_filename) != 0) {
mexErrMsgTxt("glpkcc: unable to write the problem");
longjmp (mark, -1);
}
}else{
if (!strcmp(filetype,"fixedmps")){
if (lpx_write_mps (lp, save_filename) != 0) {
mexErrMsgTxt("glpkcc: unable to write the problem");
longjmp (mark, -1);
}
}else{
if (!strcmp(filetype,"freemps")){
if (lpx_write_freemps (lp, save_filename) != 0) {
mexErrMsgTxt("glpkcc: unable to write the problem");
longjmp (mark, -1);
}
}else{// plain text
if (lpx_print_prob (lp, save_filename) != 0) {
mexErrMsgTxt("glpkcc: unable to write the problem");
longjmp (mark, -1);
}
}
}
}
}
//-- scale the problem data (if required)
if (glpIntParam[1] && (! glpIntParam[16] || lpsolver != 1))
lpx_scale_prob (lp);
//-- build advanced initial basis (if required)
if (lpsolver == 1 && ! glpIntParam[16])
lpx_adv_basis (lp);
glp_smcp sParam;
glp_init_smcp(&sParam);
//-- set control parameters
if (lpsolver==1){
//remap of control parameters for simplex method
sParam.msg_lev=glpIntParam[0]; // message level
// simplex method: primal/dual
if (glpIntParam[2]==0) sParam.meth=GLP_PRIMAL;
else sParam.meth=GLP_DUALP;
// pricing technique
if (glpIntParam[3]==0) sParam.pricing=GLP_PT_STD;
else sParam.pricing=GLP_PT_PSE;
//sParam.r_test not available
sParam.tol_bnd=glpRealParam[1]; // primal feasible tollerance
sParam.tol_dj=glpRealParam[2]; // dual feasible tollerance
sParam.tol_piv=glpRealParam[3]; // pivot tollerance
sParam.obj_ll=glpRealParam[4]; // lower limit
sParam.obj_ul=glpRealParam[5]; // upper limit
// iteration limit
if (glpIntParam[5]==-1) sParam.it_lim=INT_MAX;
else sParam.it_lim=glpIntParam[5];
// time limit
if (glpRealParam[6]==-1) sParam.tm_lim=INT_MAX;
else sParam.tm_lim=(int) glpRealParam[6];
sParam.out_frq=glpIntParam[7]; // output frequency
sParam.out_dly=(int) glpRealParam[7]; // output delay
// presolver
if (glpIntParam[16]) sParam.presolve=GLP_ON;
else sParam.presolve=GLP_OFF;
}else{
for(int i = 0; i < NIntP; i++)
lpx_set_int_parm (lp, IParam[i], glpIntParam[i]);
for (int i = 0; i < NRealP; i++)
lpx_set_real_parm (lp, RParam[i], glpRealParam[i]);
}
// Choose simplex method ('S') or interior point method ('T') to solve the problem
if (lpsolver == 1)
method = 'S';
else
method = 'T';
int errnum;
switch (method){
case 'S': {
if (isMIP){
method = 'I';
errnum = lpx_intopt (lp);
}
else{
errnum = glp_simplex(lp, &sParam);
errnum += 100; //this is to avoid ambiguity in the return codes.
}
}
break;
case 'T': errnum = lpx_interior(lp); break;
default: xassert (method != method);
}
/* errnum assumes the following results:
errnum = 0 <=> No errors
errnum = 1 <=> Iteration limit exceeded.
errnum = 2 <=> Numerical problems with basis matrix.
*/
if (errnum == LPX_E_OK || errnum==100){
// Get status and object value
if (isMIP)
{
*status = glp_mip_status (lp);
*fmin = glp_mip_obj_val (lp);
}
else
{
if (lpsolver == 1)
{
*status = glp_get_status (lp);
*fmin = glp_get_obj_val (lp);
}
else
{
*status = glp_ipt_status (lp);
*fmin = glp_ipt_obj_val (lp);
}
}
// Get optimal solution (if exists)
if (isMIP)
{
for (int i = 0; i < n; i++)
xmin[i] = glp_mip_col_val (lp, i+1);
}
else
{
/* Primal values */
for (int i = 0; i < n; i++)
{
if (lpsolver == 1)
xmin[i] = glp_get_col_prim (lp, i+1);
else
xmin[i] = glp_ipt_col_prim (lp, i+1);
}
/* Dual values */
for (int i = 0; i < m; i++)
{
if (lpsolver == 1) lambda[i] = glp_get_row_dual (lp, i+1);
else lambda[i] = glp_ipt_row_dual (lp, i+1);
}
/* Reduced costs */
for (int i = 0; i < glp_get_num_cols (lp); i++)
{
if (lpsolver == 1) redcosts[i] = glp_get_col_dual (lp, i+1);
else redcosts[i] = glp_ipt_col_dual (lp, i+1);
}
}
*time = (clock () - t_start) / CLOCKS_PER_SEC;
glp_ulong tpeak;
lib_mem_usage(NULL, NULL, NULL, &tpeak);
*mem=(double)(4294967296.0 * tpeak.hi + tpeak.lo) / (1024);
glp_delete_prob (lp);
return 0;
}
glp_delete_prob (lp);
*status = errnum;
return errnum;
}
int CConstraints::GLPK_lp(CModel* pmodel)
{
glp_prob* lp = glp_create_prob();
int iRow = (int) m_vWeights.size();
//int iCol = (int) m_iWeightLength + m_iPatternNum;
int iSize = iRow * ( m_iWeightLength + 1);
int ia[10 + iSize], ja[1 + iSize];
double ar[1 + iSize];
glp_set_prob_name(lp, "StrLP");
glp_set_obj_dir(lp, GLP_MIN);
glp_add_rows(lp, (int) m_vWeights.size());
// setup rows
for (int i = 0; i < (int) m_vWeights.size(); i ++)
{
char tmp[200];
sprintf(tmp, "cc_%d", i + 1);
glp_set_row_name(lp, i + 1, tmp);
glp_set_row_bnds(lp, i + 1, GLP_LO, m_fDistance - m_fEpsilon, 0);
}
glp_add_cols(lp, m_iWeightLength + m_iPatternNum);
for (int i = 0; i < m_iWeightLength; i ++)
{
char tmp[200];
sprintf(tmp, "w%d", i + 1);
glp_set_col_name(lp, i + 1, tmp);
glp_set_col_bnds(lp, i + 1, GLP_LO, 0, 0.0);
glp_set_obj_coef(lp, i + 1, 1.0);
}
for (int i = 0; i < m_iPatternNum; i ++)
{
char tmp[200];
sprintf(tmp, "e%d", i + 1);
glp_set_col_name(lp, m_iWeightLength + i + 1, tmp);
glp_set_col_bnds(lp, m_iWeightLength + i + 1, GLP_LO, 0, 0.0);
glp_set_obj_coef(lp, m_iWeightLength + i + 1, m_fC / m_iPatternNum);
}
int iIndex = 1;
for (int i = 0; i < (int)m_vWeights.size(); i ++)
{
double* pd = m_vWeights[i];
for (int j = 0; j < (int) m_iWeightLength; j ++)
{
ia[iIndex] = i + 1, ja[iIndex] = j + 1;
if (pmodel->m_vSign[j] <= 0)
{
ar[iIndex] = -pd[j];
}
else
{
ar[iIndex] = pd[j];
}
iIndex ++;
}
ia[iIndex] = i + 1;
ja[iIndex] = m_iWeightLength + m_vPatternIndex[i] + 1;
//ar[iIndex] = 1;
ar[iIndex] = m_vLoss[i];
iIndex ++;
}
glp_load_matrix(lp, iIndex - 1, ia, ja, ar);
glp_simplex(lp, NULL);
double z = glp_get_obj_val(lp);
fprintf(stderr, "minimal value %f \n", z);
for (int i = 0; i < m_iWeightLength; i ++)
{
double x = glp_get_col_prim(lp, i + 1);
if (pmodel->m_vSign[i] <=0)
{
pmodel->m_vWeight[i] = -x;
}
else
{
pmodel->m_vWeight[i] = x;
}
if (x != 0) fprintf(stderr, "(w%d, %f)\t", i + 1, pmodel->m_vWeight[i]);
}
for (int i = 0; i < m_iPatternNum; i ++)
{
double x = glp_get_col_prim(lp, m_iWeightLength + i + 1);
pmodel->m_vTheta[i] = x;
if (x != 0) fprintf(stderr, "(e%d, %f)\t", i + 1, pmodel->m_vTheta[i]);
}
glp_delete_prob(lp);
return 1;
}
int glpk (int sense, int n, int m, double *c, int nz, int *rn, int *cn,
double *a, double *b, char *ctype, int *freeLB, double *lb,
int *freeUB, double *ub, int *vartype, int isMIP, int lpsolver,
int save_pb, char *save_filename, char *filetype,
double *xmin, double *fmin, double *status,
double *lambda, double *redcosts, double *time, double *mem)
{
int typx = 0;
int method;
clock_t t_start = clock();
//Redirect standard output
if (glpIntParam[0] > 1) glp_term_hook (glpk_print_hook, NULL);
else glp_term_hook (NULL, NULL);
//-- Create an empty LP/MILP object
LPX *lp = lpx_create_prob ();
//-- Set the sense of optimization
if (sense == 1)
glp_set_obj_dir (lp, GLP_MIN);
else
glp_set_obj_dir (lp, GLP_MAX);
//-- Define the number of unknowns and their domains.
glp_add_cols (lp, n);
for (int i = 0; i < n; i++)
{
//-- Define type of the structural variables
if (! freeLB[i] && ! freeUB[i]) {
if ( lb[i] == ub[i] )
glp_set_col_bnds (lp, i+1, GLP_FX, lb[i], ub[i]);
else
glp_set_col_bnds (lp, i+1, GLP_DB, lb[i], ub[i]);
}
else
{
if (! freeLB[i] && freeUB[i])
glp_set_col_bnds (lp, i+1, GLP_LO, lb[i], ub[i]);
else
{
if (freeLB[i] && ! freeUB[i])
glp_set_col_bnds (lp, i+1, GLP_UP, lb[i], ub[i]);
else
glp_set_col_bnds (lp, i+1, GLP_FR, lb[i], ub[i]);
}
}
// -- Set the objective coefficient of the corresponding
// -- structural variable. No constant term is assumed.
glp_set_obj_coef(lp,i+1,c[i]);
if (isMIP)
glp_set_col_kind (lp, i+1, vartype[i]);
}
glp_add_rows (lp, m);
for (int i = 0; i < m; i++)
{
/* If the i-th row has no lower bound (types F,U), the
corrispondent parameter will be ignored.
If the i-th row has no upper bound (types F,L), the corrispondent
parameter will be ignored.
If the i-th row is of S type, the i-th LB is used, but
the i-th UB is ignored.
*/
switch (ctype[i])
{
case 'F': typx = GLP_FR; break;
// upper bound
case 'U': typx = GLP_UP; break;
// lower bound
case 'L': typx = GLP_LO; break;
// fixed constraint
case 'S': typx = GLP_FX; break;
// double-bounded variable
case 'D': typx = GLP_DB; break;
}
if ( typx == GLP_DB && -b[i] < b[i]) {
glp_set_row_bnds (lp, i+1, typx, -b[i], b[i]);
}
else if(typx == GLP_DB && -b[i] == b[i]) {
glp_set_row_bnds (lp, i+1, GLP_FX, b[i], b[i]);
}
else {
// this should be glp_set_row_bnds (lp, i+1, typx, -b[i], b[i]);
glp_set_row_bnds (lp, i+1, typx, b[i], b[i]);
}
}
// Load constraint matrix A
glp_load_matrix (lp, nz, rn, cn, a);
// Save problem
if (save_pb) {
if (!strcmp(filetype,"cplex")){
if (glp_write_lp (lp, NULL, save_filename) != 0) {
mexErrMsgTxt("glpk: unable to write the problem");
longjmp (mark, -1);
}
}else{
if (!strcmp(filetype,"fixedmps")){
if (glp_write_mps (lp, GLP_MPS_DECK, NULL, save_filename) != 0) {
mexErrMsgTxt("glpk: unable to write the problem");
longjmp (mark, -1);
}
}else{
if (!strcmp(filetype,"freemps")){
if (glp_write_mps (lp, GLP_MPS_FILE, NULL, save_filename) != 0) {
mexErrMsgTxt("glpk: unable to write the problem");
longjmp (mark, -1);
}
}else{// plain text
if (lpx_print_prob (lp, save_filename) != 0) {
mexErrMsgTxt("glpk: unable to write the problem");
longjmp (mark, -1);
}
}
}
}
}
//-- scale the problem data (if required)
if (! glpIntParam[16] || lpsolver != 1) {
switch ( glpIntParam[1] ) {
case ( 0 ): glp_scale_prob( lp, GLP_SF_SKIP ); break;
case ( 1 ): glp_scale_prob( lp, GLP_SF_GM ); break;
case ( 2 ): glp_scale_prob( lp, GLP_SF_EQ ); break;
case ( 3 ): glp_scale_prob( lp, GLP_SF_AUTO ); break;
case ( 4 ): glp_scale_prob( lp, GLP_SF_2N ); break;
default :
mexErrMsgTxt("glpk: unrecognized scaling option");
longjmp (mark, -1);
}
}
else {
/* do nothing? or unscale?
glp_unscale_prob( lp );
*/
}
//-- build advanced initial basis (if required)
if (lpsolver == 1 && ! glpIntParam[16])
glp_adv_basis (lp, 0);
glp_smcp sParam;
glp_init_smcp(&sParam);
//-- set control parameters for simplex/exact method
if (lpsolver == 1 || lpsolver == 3){
//remap of control parameters for simplex method
sParam.msg_lev=glpIntParam[0]; // message level
// simplex method: primal/dual
switch ( glpIntParam[2] ) {
case 0: sParam.meth=GLP_PRIMAL; break;
case 1: sParam.meth=GLP_DUAL; break;
case 2: sParam.meth=GLP_DUALP; break;
default:
mexErrMsgTxt("glpk: unrecognized primal/dual method");
longjmp (mark, -1);
}
// pricing technique
if (glpIntParam[3]==0) sParam.pricing=GLP_PT_STD;
else sParam.pricing=GLP_PT_PSE;
// ratio test
if (glpIntParam[20]==0) sParam.r_test = GLP_RT_STD;
else sParam.r_test=GLP_RT_HAR;
//tollerances
sParam.tol_bnd=glpRealParam[1]; // primal feasible tollerance
sParam.tol_dj=glpRealParam[2]; // dual feasible tollerance
sParam.tol_piv=glpRealParam[3]; // pivot tollerance
sParam.obj_ll=glpRealParam[4]; // lower limit
sParam.obj_ul=glpRealParam[5]; // upper limit
// iteration limit
if (glpIntParam[5]==-1) sParam.it_lim=INT_MAX;
else sParam.it_lim=glpIntParam[5];
// time limit
if (glpRealParam[6]==-1) sParam.tm_lim=INT_MAX;
else sParam.tm_lim=(int) glpRealParam[6];
sParam.out_frq=glpIntParam[7]; // output frequency
sParam.out_dly=(int) glpRealParam[7]; // output delay
// presolver
if (glpIntParam[16]) sParam.presolve=GLP_ON;
else sParam.presolve=GLP_OFF;
}else{
for(int i = 0; i < NIntP; i++) {
// skip assinging ratio test or
if ( i == 18 || i == 20) continue;
lpx_set_int_parm (lp, IParam[i], glpIntParam[i]);
}
for (int i = 0; i < NRealP; i++) {
lpx_set_real_parm (lp, RParam[i], glpRealParam[i]);
}
}
//set MIP params if MIP....
glp_iocp iParam;
glp_init_iocp(&iParam);
if ( isMIP ){
method = 'I';
switch (glpIntParam[0]) { //message level
case 0: iParam.msg_lev = GLP_MSG_OFF; break;
case 1: iParam.msg_lev = GLP_MSG_ERR; break;
case 2: iParam.msg_lev = GLP_MSG_ON; break;
case 3: iParam.msg_lev = GLP_MSG_ALL; break;
default: mexErrMsgTxt("glpk: msg_lev bad param");
}
switch (glpIntParam[14]) { //branching param
case 0: iParam.br_tech = GLP_BR_FFV; break;
case 1: iParam.br_tech = GLP_BR_LFV; break;
case 2: iParam.br_tech = GLP_BR_MFV; break;
case 3: iParam.br_tech = GLP_BR_DTH; break;
default: mexErrMsgTxt("glpk: branch bad param");
}
switch (glpIntParam[15]) { //backtracking heuristic
case 0: iParam.bt_tech = GLP_BT_DFS; break;
case 1: iParam.bt_tech = GLP_BT_BFS; break;
case 2: iParam.bt_tech = GLP_BT_BLB; break;
case 3: iParam.bt_tech = GLP_BT_BPH; break;
default: mexErrMsgTxt("glpk: backtrack bad param");
}
if ( glpRealParam[8] > 0.0 && glpRealParam[8] < 1.0 )
iParam.tol_int = glpRealParam[8]; // absolute tolorence
else
mexErrMsgTxt("glpk: tolint must be between 0 and 1");
iParam.tol_obj = glpRealParam[9]; // relative tolarence
iParam.mip_gap = glpRealParam[10]; // realative gap tolerance
// set time limit for mip
if ( glpRealParam[6] < 0.0 || glpRealParam[6] > 1e6 )
iParam.tm_lim = INT_MAX;
else
iParam.tm_lim = (int)(1000.0 * glpRealParam[6] );
// Choose Cutsets for mip
// shut all cuts off, then start over....
iParam.gmi_cuts = GLP_OFF;
iParam.mir_cuts = GLP_OFF;
iParam.cov_cuts = GLP_OFF;
iParam.clq_cuts = GLP_OFF;
switch( glpIntParam[17] ) {
case 0: break;
case 1: iParam.gmi_cuts = GLP_ON; break;
case 2: iParam.mir_cuts = GLP_ON; break;
case 3: iParam.cov_cuts = GLP_ON; break;
case 4: iParam.clq_cuts = GLP_ON; break;
case 5: iParam.clq_cuts = GLP_ON;
iParam.gmi_cuts = GLP_ON;
iParam.mir_cuts = GLP_ON;
iParam.cov_cuts = GLP_ON;
iParam.clq_cuts = GLP_ON; break;
default: mexErrMsgTxt("glpk: cutset bad param");
}
switch( glpIntParam[18] ) { // pre-processing for mip
case 0: iParam.pp_tech = GLP_PP_NONE; break;
case 1: iParam.pp_tech = GLP_PP_ROOT; break;
case 2: iParam.pp_tech = GLP_PP_ALL; break;
default: mexErrMsgTxt("glpk: pprocess bad param");
}
if (glpIntParam[16]) iParam.presolve=GLP_ON;
else iParam.presolve=GLP_OFF;
if (glpIntParam[19]) iParam.binarize = GLP_ON;
else iParam.binarize = GLP_OFF;
}
else {
/* Choose simplex method ('S')
or interior point method ('T')
or Exact method ('E')
to solve the problem */
switch (lpsolver) {
case 1: method = 'S'; break;
case 2: method = 'T'; break;
case 3: method = 'E'; break;
default:
mexErrMsgTxt("glpk: lpsolver != lpsolver");
longjmp (mark, -1);
}
}
// now run the problem...
int errnum = 0;
switch (method) {
case 'I':
errnum = glp_intopt( lp, &iParam );
errnum += 200; //this is to avoid ambiguity in the return codes.
break;
case 'S':
errnum = glp_simplex(lp, &sParam);
errnum += 100; //this is to avoid ambiguity in the return codes.
break;
case 'T':
errnum = glp_interior(lp, NULL );
errnum += 300; //this is to avoid ambiguity in the return codes.
break;
case 'E':
errnum = glp_exact(lp, &sParam);
errnum += 100; //this is to avoid ambiguity in the return codes.
break;
default: /*xassert (method != method); */
mexErrMsgTxt("glpk: method != method");
longjmp (mark, -1);
}
if (errnum==100 || errnum==200 || errnum==300 || errnum==106 || errnum==107 || errnum==108 || errnum==109 || errnum==209 || errnum==214 || errnum==308) {
// Get status and object value
if (isMIP) {
*status = glp_mip_status (lp);
*fmin = glp_mip_obj_val (lp);
}
else {
if (lpsolver == 1 || lpsolver == 3) {
*status = glp_get_status (lp);
*fmin = glp_get_obj_val (lp);
}
else {
*status = glp_ipt_status (lp);
*fmin = glp_ipt_obj_val (lp);
}
}
// Get optimal solution (if exists)
if (isMIP) {
for (int i = 0; i < n; i++)
xmin[i] = glp_mip_col_val (lp, i+1);
}
else {
/* Primal values */
for (int i = 0; i < n; i++) {
if (lpsolver == 1 || lpsolver == 3)
xmin[i] = glp_get_col_prim (lp, i+1);
else
xmin[i] = glp_ipt_col_prim (lp, i+1);
}
/* Dual values */
for (int i = 0; i < m; i++) {
if (lpsolver == 1 || lpsolver == 3)
lambda[i] = glp_get_row_dual (lp, i+1);
else
lambda[i] = glp_ipt_row_dual (lp, i+1);
}
/* Reduced costs */
for (int i = 0; i < glp_get_num_cols (lp); i++) {
if (lpsolver == 1 || lpsolver == 3)
redcosts[i] = glp_get_col_dual (lp, i+1);
else
redcosts[i] = glp_ipt_col_dual (lp, i+1);
}
}
*time = (clock () - t_start) / CLOCKS_PER_SEC;
size_t tpeak;
glp_mem_usage(NULL, NULL, NULL, &tpeak);
*mem=((double) tpeak) / (1024);
lpx_delete_prob(lp);
return 0;
}
else {
// printf("errnum is %d\n", errnum);
}
lpx_delete_prob(lp);
/* this shouldn't be nessiary with glp_deleted_prob, but try it
if we have weird behavior again... */
glp_free_env();
*status = errnum;
return errnum;
}
double c_glp_get_col_prim(glp_prob *lp, int i){
return glp_get_col_prim(lp, i);
}
static PyObject *simplex(PyObject *self, PyObject *args,
PyObject *kwrds)
{
matrix *c, *h, *b=NULL, *x=NULL, *z=NULL, *y=NULL;
PyObject *G, *A=NULL, *t=NULL;
glp_prob *lp;
glp_smcp *options = NULL;
pysmcp *smcpParm = NULL;
int m, n, p, i, j, k, nnz, nnzmax, *rn=NULL, *cn=NULL;
double *a=NULL, val;
char *kwlist[] = {"c", "G", "h", "A", "b","options", NULL};
if (!PyArg_ParseTupleAndKeywords(args, kwrds, "OOO|OOO!", kwlist, &c,
&G, &h, &A, &b,&smcp_t,&smcpParm)) return NULL;
if ((Matrix_Check(G) && MAT_ID(G) != DOUBLE) ||
(SpMatrix_Check(G) && SP_ID(G) != DOUBLE) ||
(!Matrix_Check(G) && !SpMatrix_Check(G))){
PyErr_SetString(PyExc_TypeError, "G must be a 'd' matrix");
return NULL;
}
if ((m = Matrix_Check(G) ? MAT_NROWS(G) : SP_NROWS(G)) <= 0)
err_p_int("m");
if ((n = Matrix_Check(G) ? MAT_NCOLS(G) : SP_NCOLS(G)) <= 0)
err_p_int("n");
if (!Matrix_Check(h) || h->id != DOUBLE) err_dbl_mtrx("h");
if (h->nrows != m || h->ncols != 1){
PyErr_SetString(PyExc_ValueError, "incompatible dimensions");
return NULL;
}
if (A){
if ((Matrix_Check(A) && MAT_ID(A) != DOUBLE) ||
(SpMatrix_Check(A) && SP_ID(A) != DOUBLE) ||
(!Matrix_Check(A) && !SpMatrix_Check(A))){
PyErr_SetString(PyExc_ValueError, "A must be a dense "
"'d' matrix or a general sparse matrix");
return NULL;
}
if ((p = Matrix_Check(A) ? MAT_NROWS(A) : SP_NROWS(A)) < 0)
err_p_int("p");
if ((Matrix_Check(A) ? MAT_NCOLS(A) : SP_NCOLS(A)) != n){
PyErr_SetString(PyExc_ValueError, "incompatible "
"dimensions");
return NULL;
}
}
else p = 0;
if (b && (!Matrix_Check(b) || b->id != DOUBLE)) err_dbl_mtrx("b");
if ((b && (b->nrows != p || b->ncols != 1)) || (!b && p !=0 )){
PyErr_SetString(PyExc_ValueError, "incompatible dimensions");
return NULL;
}
if(!smcpParm)
{
smcpParm = (pysmcp*)malloc(sizeof(*smcpParm));
glp_init_smcp(&(smcpParm->obj));
}
if(smcpParm)
{
Py_INCREF(smcpParm);
options = &smcpParm->obj;
options->presolve = 1;
}
lp = glp_create_prob();
glp_add_rows(lp, m+p);
glp_add_cols(lp, n);
for (i=0; i<n; i++){
glp_set_obj_coef(lp, i+1, MAT_BUFD(c)[i]);
glp_set_col_bnds(lp, i+1, GLP_FR, 0.0, 0.0);
}
for (i=0; i<m; i++)
glp_set_row_bnds(lp, i+1, GLP_UP, 0.0, MAT_BUFD(h)[i]);
for (i=0; i<p; i++)
glp_set_row_bnds(lp, i+m+1, GLP_FX, MAT_BUFD(b)[i],
MAT_BUFD(b)[i]);
nnzmax = (SpMatrix_Check(G) ? SP_NNZ(G) : m*n ) +
((A && SpMatrix_Check(A)) ? SP_NNZ(A) : p*n);
a = (double *) calloc(nnzmax+1, sizeof(double));
rn = (int *) calloc(nnzmax+1, sizeof(int));
cn = (int *) calloc(nnzmax+1, sizeof(int));
if (!a || !rn || !cn){
free(a); free(rn); free(cn); glp_delete_prob(lp);
return PyErr_NoMemory();
}
nnz = 0;
if (SpMatrix_Check(G)) {
for (j=0; j<n; j++) for (k=SP_COL(G)[j]; k<SP_COL(G)[j+1]; k++)
if ((val = SP_VALD(G)[k]) != 0.0){
a[1+nnz] = val;
rn[1+nnz] = SP_ROW(G)[k]+1;
cn[1+nnz] = j+1;
nnz++;
}
}
else for (j=0; j<n; j++) for (i=0; i<m; i++)
if ((val = MAT_BUFD(G)[i+j*m]) != 0.0){
a[1+nnz] = val;
rn[1+nnz] = i+1;
cn[1+nnz] = j+1;
nnz++;
}
if (A && SpMatrix_Check(A)){
for (j=0; j<n; j++) for (k=SP_COL(A)[j]; k<SP_COL(A)[j+1]; k++)
if ((val = SP_VALD(A)[k]) != 0.0){
a[1+nnz] = val;
rn[1+nnz] = m+SP_ROW(A)[k]+1;
cn[1+nnz] = j+1;
nnz++;
}
}
else for (j=0; j<n; j++) for (i=0; i<p; i++)
if ((val = MAT_BUFD(A)[i+j*p]) != 0.0){
a[1+nnz] = val;
rn[1+nnz] = m+i+1;
cn[1+nnz] = j+1;
nnz++;
}
glp_load_matrix(lp, nnz, rn, cn, a);
free(rn); free(cn); free(a);
if (!(t = PyTuple_New(A ? 4 : 3))){
glp_delete_prob(lp);
return PyErr_NoMemory();
}
switch (glp_simplex(lp,options)){
case 0:
x = (matrix *) Matrix_New(n,1,DOUBLE);
z = (matrix *) Matrix_New(m,1,DOUBLE);
if (A) y = (matrix *) Matrix_New(p,1,DOUBLE);
if (!x || !z || (A && !y)){
Py_XDECREF(x);
Py_XDECREF(z);
Py_XDECREF(y);
Py_XDECREF(t);
Py_XDECREF(smcpParm);
glp_delete_prob(lp);
return PyErr_NoMemory();
}
set_output_string(t,"optimal");
for (i=0; i<n; i++)
MAT_BUFD(x)[i] = glp_get_col_prim(lp, i+1);
PyTuple_SET_ITEM(t, 1, (PyObject *) x);
for (i=0; i<m; i++)
MAT_BUFD(z)[i] = -glp_get_row_dual(lp, i+1);
PyTuple_SET_ITEM(t, 2, (PyObject *) z);
if (A){
for (i=0; i<p; i++)
MAT_BUFD(y)[i] = -glp_get_row_dual(lp, m+i+1);
PyTuple_SET_ITEM(t, 3, (PyObject *) y);
}
Py_XDECREF(smcpParm);
glp_delete_prob(lp);
return (PyObject *) t;
case GLP_EBADB:
set_output_string(t,"incorrect initial basis");
break;
case GLP_ESING:
set_output_string(t,"singular initial basis matrix");
break;
case GLP_ECOND:
set_output_string(t,"ill-conditioned initial basis matrix");
break;
case GLP_EBOUND:
set_output_string(t,"incorrect bounds");
break;
case GLP_EFAIL:
set_output_string(t,"solver failure");
break;
case GLP_EOBJLL:
set_output_string(t,"objective function reached lower limit");
break;
case GLP_EOBJUL:
set_output_string(t,"objective function reached upper limit");
break;
case GLP_EITLIM:
set_output_string(t,"iteration limit exceeded");
break;
case GLP_ETMLIM:
set_output_string(t,"time limit exceeded");
break;
case GLP_ENOPFS:
set_output_string(t,"primal infeasible");
break;
case GLP_ENODFS:
set_output_string(t,"dual infeasible");
break;
default:
set_output_string(t,"unknown");
break;
}
Py_XDECREF(smcpParm);
glp_delete_prob(lp);
PyTuple_SET_ITEM(t, 1, Py_BuildValue(""));
PyTuple_SET_ITEM(t, 2, Py_BuildValue(""));
if (A) PyTuple_SET_ITEM(t, 3, Py_BuildValue(""));
return (PyObject *) t;
}
int c_simplex_sparse(int m, int n, DMAT(c), DMAT(b), DVEC(s)) {
glp_prob *lp;
lp = glp_create_prob();
glp_set_obj_dir(lp, GLP_MAX);
int i,j,k;
int tot = cr - n;
glp_add_rows(lp, m);
glp_add_cols(lp, n);
//printf("%d %d\n",m,n);
// the first n values
for (k=1;k<=n;k++) {
glp_set_obj_coef(lp, k, AT(c, k-1, 2));
//printf("%d %f\n",k,AT(c, k-1, 2));
}
int * ia = malloc((1+tot)*sizeof(int));
int * ja = malloc((1+tot)*sizeof(int));
double * ar = malloc((1+tot)*sizeof(double));
for (k=1; k<= tot; k++) {
ia[k] = rint(AT(c,k-1+n,0));
ja[k] = rint(AT(c,k-1+n,1));
ar[k] = AT(c,k-1+n,2);
//printf("%d %d %f\n",ia[k],ja[k],ar[k]);
}
glp_load_matrix(lp, tot, ia, ja, ar);
int t;
for (i=1;i<=m;i++) {
switch((int)rint(AT(b,i-1,0))) {
case 0: { t = GLP_FR; break; }
case 1: { t = GLP_LO; break; }
case 2: { t = GLP_UP; break; }
case 3: { t = GLP_DB; break; }
default: { t = GLP_FX; break; }
}
glp_set_row_bnds(lp, i, t , AT(b,i-1,1), AT(b,i-1,2));
}
for (j=1;j<=n;j++) {
switch((int)rint(AT(b,m+j-1,0))) {
case 0: { t = GLP_FR; break; }
case 1: { t = GLP_LO; break; }
case 2: { t = GLP_UP; break; }
case 3: { t = GLP_DB; break; }
default: { t = GLP_FX; break; }
}
glp_set_col_bnds(lp, j, t , AT(b,m+j-1,1), AT(b,m+j-1,2));
}
glp_term_out(0);
glp_simplex(lp, NULL);
sp[0] = glp_get_status(lp);
sp[1] = glp_get_obj_val(lp);
for (k=1; k<=n; k++) {
sp[k+1] = glp_get_col_prim(lp, k);
}
glp_delete_prob(lp);
free(ia);
free(ja);
free(ar);
return 0;
}
/*
R is the random contraint data in row major memory layout
ridx is an N array of integers
soln is an array of length (n+1) soln[n] is t (objective value)
active_constr is an N-length 0-1 array
*/
void solve_lp(int N, int n, double* R, int* ridx, double* soln, int* active_constr)
{
double tol = 1.0e-14;
int size = (N+1)*(n+1) + 1; // We add one because GLPK indexes arrays
// starting at 1 instead of 0.
glp_prob *lp;
int* ia = malloc(size * sizeof(int));
int* ja = malloc(size * sizeof(int));
double* ar = malloc(size * sizeof(double));
int i, j;
lp = glp_create_prob();
glp_set_prob_name(lp, "portfolio");
glp_set_obj_dir(lp, GLP_MAX);
glp_add_rows(lp, N+1);
// Sampled constraints are ">= 0"
for (i = 1; i <= N; i++) {
glp_set_row_bnds(lp, i, GLP_LO, 0.0, 0.0);
}
// Sum = 1 constraint
glp_set_row_name(lp, N+1, "sum");
glp_set_row_bnds(lp, N+1, GLP_FX, 1.0, 1.0);
glp_add_cols(lp, n+1);
// Nonnegative variables y
for (i = 1; i <= n; i++) {
glp_set_col_bnds(lp, i, GLP_LO, 0.0, 0.0);
glp_set_obj_coef(lp, i, 0.0);
}
// Free variable t
glp_set_col_name(lp, n+1, "t");
glp_set_col_bnds(lp, n+1, GLP_FR, 0.0, 0.0);
glp_set_obj_coef(lp, n+1, 1.0);
// for (i = 0; i < N*(n-1); i++) {
// printf("%d: %g\n", i, R[i]);
// }
int idx = 1;
// Sampled constraints
for (i = 1; i <= N; i++) {
// Uncertain assets
for (j = 1; j < n; j++) {
ia[idx] = i;
ja[idx] = j;
ar[idx] = R[ ridx[(i-1)] * (n-1) + (j-1) ];
idx += 1;
}
// Fixed return asset
ia[idx] = i;
ja[idx] = n;
ar[idx] = 1.05;
idx += 1;
// t
ia[idx] = i;
ja[idx] = n+1;
ar[idx] = -1.0;
idx += 1;
}
// Sum = 1 constraint
for (i = 1; i <= n; i++) {
ia[idx] = N+1;
ja[idx] = i;
ar[idx] = 1.0;
idx += 1;
}
// t
ia[idx] = N+1;
ja[idx] = n+1;
ar[idx] = 0.0;
idx += 1;
// for (i = 1; i < size; i++) {
// printf("%d %d %g\n", ia[i], ja[i], ar[i]);
// }
glp_load_matrix(lp, size-1, ia, ja, ar);
// glp_scale_prob(lp, GLP_SF_AUTO);
glp_smcp param;
glp_init_smcp(¶m);
param.meth = GLP_PRIMAL;
//glp_std_basis(lp);
glp_simplex(lp, ¶m);
double z = glp_get_obj_val(lp);
// printf("z = %g\n", z);
if (soln) {
for (i = 0; i < n; i++) {
double y = glp_get_col_prim(lp, i+1);
soln[i] = y;
// printf("y%d = %g\n", i, y);
}
double t = glp_get_col_prim(lp, n+1);
soln[n] = t;
// printf("t = %g\n", glp_get_col_prim(lp, n+1));
}
for (i = 1; i <= N; i++) {
double slack = glp_get_row_prim(lp, i);
active_constr[i-1] = fabs(slack) < tol ? 1 : 0;
// printf("constr%d %d\n", i, active_constr[i-1]);
}
glp_delete_prob(lp);
// glp_free_env();
free(ia);
free(ja);
free(ar);
}
int max_flow_lp(int nn, int ne, const int beg[/*1+ne*/],
const int end[/*1+ne*/], const int cap[/*1+ne*/], int s, int t,
int x[/*1+ne*/])
{ glp_prob *lp;
glp_smcp smcp;
int i, k, nz, flow, *rn, *cn;
double temp, *aa;
/* create LP problem instance */
lp = glp_create_prob();
/* create LP rows; i-th row is the conservation condition of the
* flow at i-th node, i = 1, ..., nn */
glp_add_rows(lp, nn);
for (i = 1; i <= nn; i++)
glp_set_row_bnds(lp, i, GLP_FX, 0.0, 0.0);
/* create LP columns; k-th column is the elementary flow thru
* k-th edge, k = 1, ..., ne; the last column with the number
* ne+1 is the total flow through the network, which goes along
* a dummy feedback edge from the sink to the source */
glp_add_cols(lp, ne+1);
for (k = 1; k <= ne; k++)
{ xassert(cap[k] > 0);
glp_set_col_bnds(lp, k, GLP_DB, -cap[k], +cap[k]);
}
glp_set_col_bnds(lp, ne+1, GLP_FR, 0.0, 0.0);
/* build the constraint matrix; structurally this matrix is the
* incidence matrix of the network, so each its column (including
* the last column for the dummy edge) has exactly two non-zero
* entries */
rn = xalloc(1+2*(ne+1), sizeof(int));
cn = xalloc(1+2*(ne+1), sizeof(int));
aa = xalloc(1+2*(ne+1), sizeof(double));
nz = 0;
for (k = 1; k <= ne; k++)
{ /* x[k] > 0 means the elementary flow thru k-th edge goes from
* node beg[k] to node end[k] */
nz++, rn[nz] = beg[k], cn[nz] = k, aa[nz] = -1.0;
nz++, rn[nz] = end[k], cn[nz] = k, aa[nz] = +1.0;
}
/* total flow thru the network goes from the sink to the source
* along the dummy feedback edge */
nz++, rn[nz] = t, cn[nz] = ne+1, aa[nz] = -1.0;
nz++, rn[nz] = s, cn[nz] = ne+1, aa[nz] = +1.0;
/* check the number of non-zero entries */
xassert(nz == 2*(ne+1));
/* load the constraint matrix into the LP problem object */
glp_load_matrix(lp, nz, rn, cn, aa);
xfree(rn);
xfree(cn);
xfree(aa);
/* objective function is the total flow through the network to
* be maximized */
glp_set_obj_dir(lp, GLP_MAX);
glp_set_obj_coef(lp, ne + 1, 1.0);
/* solve LP instance with the (primal) simplex method */
glp_term_out(0);
glp_adv_basis(lp, 0);
glp_term_out(1);
glp_init_smcp(&smcp);
smcp.msg_lev = GLP_MSG_ON;
smcp.out_dly = 5000;
xassert(glp_simplex(lp, &smcp) == 0);
xassert(glp_get_status(lp) == GLP_OPT);
/* obtain optimal elementary flows thru edges of the network */
/* (note that the constraint matrix is unimodular and the data
* are integral, so all elementary flows in basic solution should
* also be integral) */
for (k = 1; k <= ne; k++)
{ temp = glp_get_col_prim(lp, k);
x[k] = (int)floor(temp + .5);
xassert(fabs(x[k] - temp) <= 1e-6);
}
/* obtain the maximum flow thru the original network which is the
* flow thru the dummy feedback edge */
temp = glp_get_col_prim(lp, ne+1);
flow = (int)floor(temp + .5);
xassert(fabs(flow - temp) <= 1e-6);
/* delete LP problem instance */
glp_delete_prob(lp);
/* return to the calling program */
return flow;
}
double NUMlinprog_getPrimalValue (NUMlinprog me, long ivar) {
return glp_get_col_prim (my linearProgram, ivar);
}