Method Go

libs/search/streets_matcher.cpp:140–184

static

// static
void StreetsMatcher::Go(BaseContext const & ctx, CBV const & candidates, FeaturesFilter const & filter,
                        QueryParams const & params, vector<Prediction> & predictions)
{
  predictions.clear();
  FindStreets(ctx, candidates, filter, params, predictions);

  if (predictions.empty())
    return;

  // Remove predictions with the same m_hash (features) and token range.
  base::SortUnique(predictions, &LessByHashAndRange, &EqualsByHashAndRange);

  // We should distinguish predictions with the same m_hash (features) but different range and
  // m_withMisprints. For example, for "Paramount dive" we will have two parses:
  //
  // STREET       UNUSED (can be matched to poi later)
  // Paramount   dive
  //
  // STREET       STREET ("drive" with misprints)
  // Paramount    dive
  //
  // The parses will have the same features and hash but we need both of them.
  //
  // We also need to distinguish predictions with the same m_hash (features) and m_withMisprints but
  // different range. For example:
  //
  // STREET STREET STREET  STREET
  // 8      March  street, 8
  //
  // STREET STREET STREET  UNUSED (can be matched to house number later)
  // 8      March  street, 8
  //
  // Predictions have the same m_hash (features) and m_withMisprints but lead to different parses.
  //
  // That's why we need all predictions here.

  sort(predictions.begin(), predictions.end(),
       [](Prediction const & l, Prediction const & r) { return l.IsBetter(r); });

  // I suppose, it was made to avoid matching by *very* common tokens (like 'street' only).
  size_t constexpr kMaxNumOfImprobablePredictions = 3;
  double constexpr kTailProbability = 0.05;
  while (predictions.size() > kMaxNumOfImprobablePredictions && predictions.back().m_prob < kTailProbability)
    predictions.pop_back();
}

// static
void StreetsMatcher::FindStreets(BaseContext const & ctx, CBV const & candidates, FeaturesFilter const & filter,