Method BeforeTrain

src/treelearner/serial_tree_learner.cpp:321–398

}

void SerialTreeLearner::BeforeTrain() {
  // reset histogram pool
  histogram_pool_.ResetMap();

  if (config_->feature_fraction < 1.0f) {
    is_feature_used_ = GetUsedFeatures(true);
  } else {
    #pragma omp parallel for schedule(static, 512) if (num_features_ >= 1024)
    for (int i = 0; i < num_features_; ++i) {
      is_feature_used_[i] = 1;
    }
  }

  // initialize data partition
  data_partition_->Init();

  // reset the splits for leaves
  for (int i = 0; i < config_->num_leaves; ++i) {
    best_split_per_leaf_[i].Reset();
  }

  // Sumup for root
  if (data_partition_->leaf_count(0) == num_data_) {
    // use all data
    smaller_leaf_splits_->Init(gradients_, hessians_);

  } else {
    // use bagging, only use part of data
    smaller_leaf_splits_->Init(0, data_partition_.get(), gradients_, hessians_);
  }

  larger_leaf_splits_->Init();

  // if has ordered bin, need to initialize the ordered bin
  if (has_ordered_bin_) {
    #ifdef TIMETAG
    auto start_time = std::chrono::steady_clock::now();
    #endif
    if (data_partition_->leaf_count(0) == num_data_) {
      // use all data, pass nullptr
      OMP_INIT_EX();
      #pragma omp parallel for schedule(static)
      for (int i = 0; i < static_cast<int>(ordered_bin_indices_.size()); ++i) {
        OMP_LOOP_EX_BEGIN();
        ordered_bins_[ordered_bin_indices_[i]]->Init(nullptr, config_->num_leaves);
        OMP_LOOP_EX_END();
      }
      OMP_THROW_EX();
    } else {
      // bagging, only use part of data

      // mark used data
      const data_size_t* indices = data_partition_->indices();
      data_size_t begin = data_partition_->leaf_begin(0);
      data_size_t end = begin + data_partition_->leaf_count(0);
      #pragma omp parallel for schedule(static, 512) if (end - begin >= 1024)
      for (data_size_t i = begin; i < end; ++i) {
        is_data_in_leaf_[indices[i]] = 1;