Eigen::VectorXd _inner_search(const F& f, size_t depth, const Eigen::VectorXd& current) const
{
size_t dim = current.size();
double step_size = 1.0 / (double)Params::opt_gridsearch::bins();
double upper_lim = 1.0 + step_size;
double best_fit = -std::numeric_limits<double>::max();
Eigen::VectorXd current_result(dim);
for (double x = 0; x < upper_lim; x += step_size) {
Eigen::VectorXd new_point = current;
new_point[depth] = x;
double val;
if (depth == dim - 1) {
val = eval(f, new_point);
if (val > best_fit) {
best_fit = val;
current_result = new_point;
}
}
else {
Eigen::VectorXd temp_result = _inner_search(f, depth + 1, new_point);
val = eval(f, temp_result);
if (val > best_fit) {
best_fit = val;
current_result = temp_result;
}
}
}
return current_result;
}
Eigen::VectorXd explore(int dim,
const AcquisitionFunction& acqui,
const Eigen::VectorXd& current,
Eigen::VectorXd& result) const {
double best_fit = -std::numeric_limits<double>::max();
Eigen::VectorXd current_result(result.size());
for (double x = 0; x <= 1; x += 1 / (double)nb_pts) {
Eigen::VectorXd point = current;
point[dim] = x;
double val;
if (dim == current.size() - 1) {
val = acqui(point);
if (val > best_fit) {
best_fit = val;
current_result = point;
}
} else {
Eigen::VectorXd temp_result = current_result;
std::tie(temp_result, val) = explore(dim + 1, acqui, point, temp_result);
if (val > best_fit) {
best_fit = val;
current_result = temp_result;
}
}
}
return current_result;
}
