int main() {
int racer_num = 5;
Racer rcs[] = {
{ 1, 100 },
{ 2, 10 },
{ 3, 4 },
{ 5, 6 },
};
vector<Racer> racers(rcs, rcs+ArraySize(rcs));
// for (int i=0; i<racer_num; ++i) {
// racers[i].beg = rand();
// racers[i].end = racers[i].beg + rand();
// }
vector<int> scores = ScoreHelper(racers).GetScores();
return 0;
}
// Note: ignore parallel paramter, only applies to other IScorers
ScoreResult Tree::Score(const bkp::MaskedVector<const HiggsCsvRow>& data, bool parallel) {
int data_size = static_cast<int>(data.size());
std::vector<double> s_scores(data_size, std::numeric_limits<double>::quiet_NaN());
std::vector<double> b_scores(data_size, std::numeric_limits<double>::quiet_NaN());
bkp::MaskedVector<int> return_indices(
std::vector<int>(
boost::counting_iterator<int>(0),
boost::counting_iterator<int>(data_size)
)
);
// construct filter such that filter[i] is true iff data has no NaN values in any of
// our target_feature columns (don't care about NaNs in other columns)
std::vector<bool> filter = HasNan(data, *target_features_);
filter.flip();
// use local std::vectors as backing stores for some MaskingVectors. ScoreHelper
// will then populate our vectors by populating the MaskingVector and their
// MaskingVector::Slices (implicitly ignoring rows with NaN's, which will be
// handled by other trees).
//
// Normally it would be dangerous to create a shared_ptr to a local variable like
// this. Either the shared_ptr will go out of scope and try to delete the
// local variable (crash and/or UB) or the local variable will go out of scope with the
// shared_ptr still pointing to it (dangling ptr). In this case we can see that
// the local variable will outlive the shared_ptrs (they will both live until the
// end of this method, although the contents of the local vectors will be std::moved
// to a value-returned Score object at the end of the method). So, all we need to do
// is remove the shared_ptr's deleter (done in ShareLocal) and we're good to go.
auto filtered_return_indices = return_indices.Filter(filter);
ScoreHelper(data,
filtered_return_indices.MakeSlice(),
s_scores,
b_scores);
return hrf::ScoreResult(std::move(s_scores), std::move(b_scores));
}
