xgboost/src/predictor/cpu_predictor.cc

/*!
 * Copyright by Contributors 2017
 */
#include <xgboost/predictor.h>
#include <xgboost/tree_model.h>
#include <xgboost/tree_updater.h>
#include "dmlc/logging.h"
#include "../common/host_device_vector.h"

namespace xgboost {
namespace predictor {

DMLC_REGISTRY_FILE_TAG(cpu_predictor);

class CPUPredictor : public Predictor {
 protected:
  static bst_float PredValue(const RowBatch::Inst& inst,
                             const std::vector<std::unique_ptr<RegTree>>& trees,
                             const std::vector<int>& tree_info, int bst_group,
                             unsigned root_index, RegTree::FVec* p_feats,
                             unsigned tree_begin, unsigned tree_end) {
    bst_float psum = 0.0f;
    p_feats->Fill(inst);
    for (size_t i = tree_begin; i < tree_end; ++i) {
      if (tree_info[i] == bst_group) {
        int tid = trees[i]->GetLeafIndex(*p_feats, root_index);
        psum += (*trees[i])[tid].leaf_value();
      }
    }
    p_feats->Drop(inst);
    return psum;
  }

  // init thread buffers
  inline void InitThreadTemp(int nthread, int num_feature) {
    int prev_thread_temp_size = thread_temp.size();
    if (prev_thread_temp_size < nthread) {
      thread_temp.resize(nthread, RegTree::FVec());
      for (int i = prev_thread_temp_size; i < nthread; ++i) {
        thread_temp[i].Init(num_feature);
      }
    }
  }
  inline void PredLoopSpecalize(DMatrix* p_fmat,
                                std::vector<bst_float>* out_preds,
                                const gbm::GBTreeModel& model, int num_group,
                                unsigned tree_begin, unsigned tree_end) {
    const MetaInfo& info = p_fmat->info();
    const int nthread = omp_get_max_threads();
    InitThreadTemp(nthread, model.param.num_feature);
    std::vector<bst_float>& preds = *out_preds;
    CHECK_EQ(model.param.size_leaf_vector, 0)
        << "size_leaf_vector is enforced to 0 so far";
    CHECK_EQ(preds.size(), p_fmat->info().num_row * num_group);
    // start collecting the prediction
    dmlc::DataIter<RowBatch>* iter = p_fmat->RowIterator();
    iter->BeforeFirst();
    while (iter->Next()) {
      const RowBatch& batch = iter->Value();
      // parallel over local batch
      const int K = 8;
      const bst_omp_uint nsize = static_cast<bst_omp_uint>(batch.size);
      const bst_omp_uint rest = nsize % K;
#pragma omp parallel for schedule(static)
      for (bst_omp_uint i = 0; i < nsize - rest; i += K) {
        const int tid = omp_get_thread_num();
        RegTree::FVec& feats = thread_temp[tid];
        int64_t ridx[K];
        RowBatch::Inst inst[K];
        for (int k = 0; k < K; ++k) {
          ridx[k] = static_cast<int64_t>(batch.base_rowid + i + k);
        }
        for (int k = 0; k < K; ++k) {
          inst[k] = batch[i + k];
        }
        for (int k = 0; k < K; ++k) {
          for (int gid = 0; gid < num_group; ++gid) {
            const size_t offset = ridx[k] * num_group + gid;
            preds[offset] += this->PredValue(
                inst[k], model.trees, model.tree_info, gid,
                info.GetRoot(ridx[k]), &feats, tree_begin, tree_end);
          }
        }
      }
      for (bst_omp_uint i = nsize - rest; i < nsize; ++i) {
        RegTree::FVec& feats = thread_temp[0];
        const int64_t ridx = static_cast<int64_t>(batch.base_rowid + i);
        const RowBatch::Inst inst = batch[i];
        for (int gid = 0; gid < num_group; ++gid) {
          const size_t offset = ridx * num_group + gid;
          preds[offset] +=
              this->PredValue(inst, model.trees, model.tree_info, gid,
                              info.GetRoot(ridx), &feats, tree_begin, tree_end);
        }
      }
    }
  }

  void PredLoopInternal(DMatrix* dmat, std::vector<bst_float>* out_preds,
                        const gbm::GBTreeModel& model, int tree_begin,
                        unsigned ntree_limit) {
    // TODO(Rory): Check if this specialisation actually improves performance
    if (model.param.num_output_group == 1) {
      PredLoopSpecalize(dmat, out_preds, model, 1, tree_begin, ntree_limit);
    } else {
      PredLoopSpecalize(dmat, out_preds, model, model.param.num_output_group,
                        tree_begin, ntree_limit);
    }
  }

 public:
  void PredictBatch(DMatrix* dmat, HostDeviceVector<bst_float>* out_preds,
                    const gbm::GBTreeModel& model, int tree_begin,
                    unsigned ntree_limit = 0) override {
    PredictBatch(dmat, &out_preds->data_h(), model, tree_begin, ntree_limit);
  }

  void PredictBatch(DMatrix* dmat, std::vector<bst_float>* out_preds,
                    const gbm::GBTreeModel& model, int tree_begin,
                    unsigned ntree_limit = 0) override {
    if (this->PredictFromCache(dmat, out_preds, model, ntree_limit)) {
      return;
    }

    this->InitOutPredictions(dmat->info(), out_preds, model);

    ntree_limit *= model.param.num_output_group;
    if (ntree_limit == 0 || ntree_limit > model.trees.size()) {
      ntree_limit = static_cast<unsigned>(model.trees.size());
    }

    this->PredLoopInternal(dmat, out_preds, model, tree_begin, ntree_limit);
  }

  void UpdatePredictionCache(
      const gbm::GBTreeModel& model,
      std::vector<std::unique_ptr<TreeUpdater>>* updaters,
      int num_new_trees) override {
    int old_ntree = model.trees.size() - num_new_trees;
    // update cache entry
    for (auto& kv : cache_) {
      PredictionCacheEntry& e = kv.second;

      if (e.predictions.size() == 0) {
        InitOutPredictions(e.data->info(), &(e.predictions), model);
        PredLoopInternal(e.data.get(), &(e.predictions), model, 0,
                         model.trees.size());
      } else if (model.param.num_output_group == 1 && updaters->size() > 0 &&
                 num_new_trees == 1 &&
                 updaters->back()->UpdatePredictionCache(e.data.get(),
                                                         &(e.predictions))) {
        {}  // do nothing
      } else {
        PredLoopInternal(e.data.get(), &(e.predictions), model, old_ntree,
                         model.trees.size());
      }
    }
  }

  void PredictInstance(const SparseBatch::Inst& inst,
                       std::vector<bst_float>* out_preds,
                       const gbm::GBTreeModel& model, unsigned ntree_limit,
                       unsigned root_index) override {
    if (thread_temp.size() == 0) {
      thread_temp.resize(1, RegTree::FVec());
      thread_temp[0].Init(model.param.num_feature);
    }
    ntree_limit *= model.param.num_output_group;
    if (ntree_limit == 0 || ntree_limit > model.trees.size()) {
      ntree_limit = static_cast<unsigned>(model.trees.size());
    }
    out_preds->resize(model.param.num_output_group *
                      (model.param.size_leaf_vector + 1));
    // loop over output groups
    for (int gid = 0; gid < model.param.num_output_group; ++gid) {
      (*out_preds)[gid] =
          PredValue(inst, model.trees, model.tree_info, gid, root_index,
                    &thread_temp[0], 0, ntree_limit) +
          model.base_margin;
    }
  }
  void PredictLeaf(DMatrix* p_fmat, std::vector<bst_float>* out_preds,
                   const gbm::GBTreeModel& model, unsigned ntree_limit) override {
    const int nthread = omp_get_max_threads();
    InitThreadTemp(nthread, model.param.num_feature);
    const MetaInfo& info = p_fmat->info();
    // number of valid trees
    ntree_limit *= model.param.num_output_group;
    if (ntree_limit == 0 || ntree_limit > model.trees.size()) {
      ntree_limit = static_cast<unsigned>(model.trees.size());
    }
    std::vector<bst_float>& preds = *out_preds;
    preds.resize(info.num_row * ntree_limit);
    // start collecting the prediction
    dmlc::DataIter<RowBatch>* iter = p_fmat->RowIterator();
    iter->BeforeFirst();
    while (iter->Next()) {
      const RowBatch& batch = iter->Value();
      // parallel over local batch
      const bst_omp_uint nsize = static_cast<bst_omp_uint>(batch.size);
#pragma omp parallel for schedule(static)
      for (bst_omp_uint i = 0; i < nsize; ++i) {
        const int tid = omp_get_thread_num();
        size_t ridx = static_cast<size_t>(batch.base_rowid + i);
        RegTree::FVec& feats = thread_temp[tid];
        feats.Fill(batch[i]);
        for (unsigned j = 0; j < ntree_limit; ++j) {
          int tid = model.trees[j]->GetLeafIndex(feats, info.GetRoot(ridx));
          preds[ridx * ntree_limit + j] = static_cast<bst_float>(tid);
        }
        feats.Drop(batch[i]);
      }
    }
  }

  void PredictContribution(DMatrix* p_fmat, std::vector<bst_float>* out_contribs,
                           const gbm::GBTreeModel& model, unsigned ntree_limit,
                           bool approximate,
                           int condition,
                           unsigned condition_feature) override {
    const int nthread = omp_get_max_threads();
    InitThreadTemp(nthread,  model.param.num_feature);
    const MetaInfo& info = p_fmat->info();
    // number of valid trees
    ntree_limit *= model.param.num_output_group;
    if (ntree_limit == 0 || ntree_limit > model.trees.size()) {
      ntree_limit = static_cast<unsigned>(model.trees.size());
    }
    const int ngroup = model.param.num_output_group;
    size_t ncolumns = model.param.num_feature + 1;
    // allocate space for (number of features + bias) times the number of rows
    std::vector<bst_float>& contribs = *out_contribs;
    contribs.resize(info.num_row * ncolumns * model.param.num_output_group);
    // make sure contributions is zeroed, we could be reusing a previously
    // allocated one
    std::fill(contribs.begin(), contribs.end(), 0);
    // initialize tree node mean values
    #pragma omp parallel for schedule(static)
    for (bst_omp_uint i = 0; i < ntree_limit; ++i) {
      model.trees[i]->FillNodeMeanValues();
    }
    // start collecting the contributions
    dmlc::DataIter<RowBatch>* iter = p_fmat->RowIterator();
    const std::vector<bst_float>& base_margin = info.base_margin;
    iter->BeforeFirst();
    while (iter->Next()) {
      const RowBatch& batch = iter->Value();
      // parallel over local batch
      const bst_omp_uint nsize = static_cast<bst_omp_uint>(batch.size);
#pragma omp parallel for schedule(static)
      for (bst_omp_uint i = 0; i < nsize; ++i) {
        size_t row_idx = static_cast<size_t>(batch.base_rowid + i);
        unsigned root_id = info.GetRoot(row_idx);
        RegTree::FVec& feats = thread_temp[omp_get_thread_num()];
        // loop over all classes
        for (int gid = 0; gid < ngroup; ++gid) {
          bst_float* p_contribs =
              &contribs[(row_idx * ngroup + gid) * ncolumns];
          feats.Fill(batch[i]);
          // calculate contributions
          for (unsigned j = 0; j < ntree_limit; ++j) {
            if (model.tree_info[j] != gid) {
              continue;
            }
            if (!approximate) {
              model.trees[j]->CalculateContributions(feats, root_id, p_contribs,
                                                     condition, condition_feature);
            } else {
              model.trees[j]->CalculateContributionsApprox(feats, root_id, p_contribs);
            }
          }
          feats.Drop(batch[i]);
          // add base margin to BIAS
          if (base_margin.size() != 0) {
            p_contribs[ncolumns - 1] += base_margin[row_idx * ngroup + gid];
          } else {
            p_contribs[ncolumns - 1] += model.base_margin;
          }
        }
      }
    }
  }

  void PredictInteractionContributions(DMatrix* p_fmat, std::vector<bst_float>* out_contribs,
                                       const gbm::GBTreeModel& model, unsigned ntree_limit,
                                       bool approximate) override {
    const MetaInfo& info = p_fmat->info();
    const int ngroup = model.param.num_output_group;
    size_t ncolumns = model.param.num_feature;
    const unsigned row_chunk = ngroup * (ncolumns + 1) * (ncolumns + 1);
    const unsigned mrow_chunk = (ncolumns + 1) * (ncolumns + 1);
    const unsigned crow_chunk = ngroup * (ncolumns + 1);

    // allocate space for (number of features^2) times the number of rows and tmp off/on contribs
    std::vector<bst_float>& contribs = *out_contribs;
    contribs.resize(info.num_row * ngroup * (ncolumns + 1) * (ncolumns + 1));
    std::vector<bst_float> contribs_off(info.num_row * ngroup * (ncolumns + 1));
    std::vector<bst_float> contribs_on(info.num_row * ngroup * (ncolumns + 1));
    std::vector<bst_float> contribs_diag(info.num_row * ngroup * (ncolumns + 1));

    // Compute the difference in effects when conditioning on each of the features on and off
    // see: Axiomatic characterizations of probabilistic and
    //      cardinal-probabilistic interaction indices
    PredictContribution(p_fmat, &contribs_diag, model, ntree_limit, approximate, 0, 0);
    for (size_t i = 0; i < ncolumns + 1; ++i) {
      PredictContribution(p_fmat, &contribs_off, model, ntree_limit, approximate, -1, i);
      PredictContribution(p_fmat, &contribs_on, model, ntree_limit, approximate, 1, i);

      for (size_t j = 0; j < info.num_row; ++j) {
        for (int l = 0; l < ngroup; ++l) {
          const unsigned o_offset = j * row_chunk + l * mrow_chunk + i * (ncolumns + 1);
          const unsigned c_offset = j * crow_chunk + l * (ncolumns + 1);
          contribs[o_offset + i] = 0;
          for (size_t k = 0; k < ncolumns + 1; ++k) {
            // fill in the diagonal with additive effects, and off-diagonal with the interactions
            if (k == i) {
              contribs[o_offset + i] += contribs_diag[c_offset + k];
            } else {
              contribs[o_offset + k] = (contribs_on[c_offset + k] - contribs_off[c_offset + k])/2.0;
              contribs[o_offset + i] -= contribs[o_offset + k];
            }
          }
        }
      }
    }
  }
  std::vector<RegTree::FVec> thread_temp;
};

XGBOOST_REGISTER_PREDICTOR(CPUPredictor, "cpu_predictor")
    .describe("Make predictions using CPU.")
    .set_body([]() { return new CPUPredictor(); });
}  // namespace predictor
}  // namespace xgboost