void QpToNlp::init(const Dict& opts) {
// Initialize the base classes
Qpsol::init(opts);
// Default options
string nlpsol_plugin;
Dict nlpsol_options;
// Read user options
for (auto&& op : opts) {
if (op.first=="nlpsol") {
nlpsol_plugin = op.second.to_string();
} else if (op.first=="nlpsol_options") {
nlpsol_options = op.second;
}
}
// Create a symbolic matrix for the decision variables
SX X = SX::sym("X", n_, 1);
// Parameters to the problem
SX H = SX::sym("H", sparsity_in(QPSOL_H));
SX G = SX::sym("G", sparsity_in(QPSOL_G));
SX A = SX::sym("A", sparsity_in(QPSOL_A));
// Put parameters in a vector
std::vector<SX> par;
par.push_back(H.nonzeros());
par.push_back(G.nonzeros());
par.push_back(A.nonzeros());
// The nlp looks exactly like a mathematical description of the NLP
SXDict nlp = {{"x", X}, {"p", vertcat(par)},
{"f", mtimes(G.T(), X) + 0.5*mtimes(mtimes(X.T(), H), X)},
{"g", mtimes(A, X)}};
// Create an Nlpsol instance
casadi_assert_message(!nlpsol_plugin.empty(), "'nlpsol' option has not been set");
solver_ = nlpsol("nlpsol", nlpsol_plugin, nlp, nlpsol_options);
alloc(solver_);
// Allocate storage for NLP solver parameters
alloc_w(solver_.nnz_in(NLPSOL_P), true);
}
void Callback::eval(const double** arg, double** res, int* iw, double* w, int mem) {
// Allocate input matrices
int n_in = this->n_in();
std::vector<DM> argv(n_in);
for (int i=0; i<n_in; ++i) {
argv[i] = DM(sparsity_in(i));
casadi_copy(arg[i], argv[i].nnz(), argv[i].ptr());
}
// Evaluate
std::vector<DM> resv = eval(argv);
// Get the outputs
int n_out = this->n_out();
for (int i=0; i<n_out; ++i) {
casadi_copy(resv[i].ptr(), resv[i].nnz(), res[i]);
}
}
void GurobiInterface::
eval(void* mem, const double** arg, double** res, int* iw, double* w) const {
auto m = static_cast<GurobiMemory*>(mem);
// Inputs
const double *h=arg[CONIC_H],
*g=arg[CONIC_G],
*a=arg[CONIC_A],
*lba=arg[CONIC_LBA],
*uba=arg[CONIC_UBA],
*lbx=arg[CONIC_LBX],
*ubx=arg[CONIC_UBX],
*x0=arg[CONIC_X0],
*lam_x0=arg[CONIC_LAM_X0];
// Outputs
double *x=res[CONIC_X],
*cost=res[CONIC_COST],
*lam_a=res[CONIC_LAM_A],
*lam_x=res[CONIC_LAM_X];
// Temporary memory
double *val=w; w+=nx_;
int *ind=iw; iw+=nx_;
int *ind2=iw; iw+=nx_;
int *tr_ind=iw; iw+=nx_;
// Greate an empty model
GRBmodel *model = 0;
try {
int flag = GRBnewmodel(m->env, &model, name_.c_str(), 0, 0, 0, 0, 0, 0);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
// Add variables
for (int i=0; i<nx_; ++i) {
// Get bounds
double lb = lbx ? lbx[i] : 0., ub = ubx ? ubx[i] : 0.;
if (isinf(lb)) lb = -GRB_INFINITY;
if (isinf(ub)) ub = GRB_INFINITY;
// Get variable type
char vtype;
if (!vtype_.empty()) {
// Explicitly set 'vtype' takes precedence
vtype = vtype_.at(i);
} else if (!discrete_.empty() && discrete_.at(i)) {
// Variable marked as discrete (integer or binary)
vtype = lb==0 && ub==1 ? GRB_BINARY : GRB_INTEGER;
} else {
// Continious variable
vtype = GRB_CONTINUOUS;
}
// Pass to model
flag = GRBaddvar(model, 0, 0, 0, g ? g[i] : 0., lb, ub, vtype, 0);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
}
flag = GRBupdatemodel(model);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
// Add quadratic terms
const int *H_colind=sparsity_in(CONIC_H).colind(), *H_row=sparsity_in(CONIC_H).row();
for (int i=0; i<nx_; ++i) {
// Quadratic term nonzero indices
int numqnz = H_colind[1]-H_colind[0];
casadi_copy(H_row, numqnz, ind);
H_colind++;
H_row += numqnz;
// Corresponding column
casadi_fill(ind2, numqnz, i);
// Quadratic term nonzeros
if (h) {
casadi_copy(h, numqnz, val);
casadi_scal(numqnz, 0.5, val);
h += numqnz;
} else {
casadi_fill(val, numqnz, 0.);
}
// Pass to model
flag = GRBaddqpterms(model, numqnz, ind, ind2, val);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
}
// Add constraints
const int *A_colind=sparsity_in(CONIC_A).colind(), *A_row=sparsity_in(CONIC_A).row();
casadi_copy(A_colind, nx_, tr_ind);
for (int i=0; i<na_; ++i) {
// Get bounds
double lb = lba ? lba[i] : 0., ub = uba ? uba[i] : 0.;
// if (isinf(lb)) lb = -GRB_INFINITY;
// if (isinf(ub)) ub = GRB_INFINITY;
// Constraint nonzeros
int numnz = 0;
for (int j=0; j<nx_; ++j) {
if (tr_ind[j]<A_colind[j+1] && A_row[tr_ind[j]]==i) {
ind[numnz] = j;
val[numnz] = a ? a[tr_ind[j]] : 0;
numnz++;
tr_ind[j]++;
}
}
// Pass to model
if (isinf(lb)) {
if (isinf(ub)) {
// Neither upper or lower bounds, skip
} else {
// Only upper bound
flag = GRBaddconstr(model, numnz, ind, val, GRB_LESS_EQUAL, ub, 0);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
}
} else {
if (isinf(ub)) {
// Only lower bound
flag = GRBaddconstr(model, numnz, ind, val, GRB_GREATER_EQUAL, lb, 0);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
} else if (lb==ub) {
// Upper and lower bounds equal
flag = GRBaddconstr(model, numnz, ind, val, GRB_EQUAL, lb, 0);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
} else {
// Both upper and lower bounds
flag = GRBaddrangeconstr(model, numnz, ind, val, lb, ub, 0);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
}
}
}
// Solve the optimization problem
flag = GRBoptimize(model);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
int optimstatus;
flag = GRBgetintattr(model, GRB_INT_ATTR_STATUS, &optimstatus);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
// Get the objective value, if requested
if (cost) {
flag = GRBgetdblattr(model, GRB_DBL_ATTR_OBJVAL, cost);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
}
// Get the optimal solution, if requested
if (x) {
flag = GRBgetdblattrarray(model, GRB_DBL_ATTR_X, 0, nx_, x);
casadi_assert_message(!flag, GRBgeterrormsg(m->env));
}
// Free memory
GRBfreemodel(model);
} catch (...) {
// Free memory
if (model) GRBfreemodel(model);
throw;
}
}
