Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GPU-Plugin] Add GPU accelerated prediction (#2593)

Browse Source 
* [GPU-Plugin] Add GPU accelerated prediction

* Improve allocation message

* Update documentation

* Resolve linker error for predictor

* Add unit tests
This commit is contained in:
Rory Mitchell
2017-08-16 12:31:59 +12:00
committed by GitHub
parent 71e5e622b1
commit ef23e424f1
25 changed files with 876 additions and 203 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
src/predictor/cpu_predictor.cc
View File

@@ -9,6 +9,8 @@
namespace xgboost {
namespace predictor {
DMLC_REGISTRY_FILE_TAG(cpu_predictor);
class CPUPredictor : public Predictor {
protected:
static bst_float PredValue(const RowBatch::Inst& inst,
@@ -28,19 +30,6 @@ class CPUPredictor : public Predictor {
return psum;
}
void InitOutPredictions(const MetaInfo& info,
std::vector<bst_float>* out_preds,
const gbm::GBTreeModel& model) const {
size_t n = model.param.num_output_group * info.num_row;
const std::vector<bst_float>& base_margin = info.base_margin;
out_preds->resize(n);
if (base_margin.size() != 0) {
CHECK_EQ(out_preds->size(), n);
std::copy(base_margin.begin(), base_margin.end(), out_preds->begin());
} else {
std::fill(out_preds->begin(), out_preds->end(), model.base_margin);
}
}
// init thread buffers
inline void InitThreadTemp(int nthread, int num_feature) {
int prev_thread_temp_size = thread_temp.size();
@@ -106,33 +95,6 @@ class CPUPredictor : public Predictor {
}
}
/**
* \fn bool PredictFromCache(DMatrix* dmat, std::vector<bst_float>*
* out_preds, const gbm::GBTreeModel& model, unsigned ntree_limit = 0)
*
* \brief Attempt to predict from cache.
*
* \return True if it succeeds, false if it fails.
*/
bool PredictFromCache(DMatrix* dmat, std::vector<bst_float>* out_preds,
const gbm::GBTreeModel& model,
unsigned ntree_limit = 0) {
if (ntree_limit == 0 ||
ntree_limit * model.param.num_output_group >= model.trees.size()) {
auto it = cache_.find(dmat);
if (it != cache_.end()) {
std::vector<bst_float>& y = it->second.predictions;
if (y.size() != 0) {
out_preds->resize(y.size());
std::copy(y.begin(), y.end(), out_preds->begin());
return true;
}
}
}
return false;
}
void PredLoopInternal(DMatrix* dmat, std::vector<bst_float>* out_preds,
const gbm::GBTreeModel& model, int tree_begin,
unsigned ntree_limit) {

46
src/predictor/predictor.cc
View File

@@ -8,13 +8,47 @@ namespace dmlc {
DMLC_REGISTRY_ENABLE(::xgboost::PredictorReg);
} // namespace dmlc
namespace xgboost {
void Predictor::InitCache(const std::vector<std::shared_ptr<DMatrix> >& cache) {
void Predictor::Init(
const std::vector<std::pair<std::string, std::string>>& cfg,
const std::vector<std::shared_ptr<DMatrix>>& cache) {
for (const std::shared_ptr<DMatrix>& d : cache) {
PredictionCacheEntry e;
e.data = d;
cache_[d.get()] = std::move(e);
}
}
bool Predictor::PredictFromCache(DMatrix* dmat,
std::vector<bst_float>* out_preds,
const gbm::GBTreeModel& model,
unsigned ntree_limit) {
if (ntree_limit == 0 ||
ntree_limit * model.param.num_output_group >= model.trees.size()) {
auto it = cache_.find(dmat);
if (it != cache_.end()) {
std::vector<bst_float>& y = it->second.predictions;
if (y.size() != 0) {
out_preds->resize(y.size());
std::copy(y.begin(), y.end(), out_preds->begin());
return true;
}
}
}
return false;
}
void Predictor::InitOutPredictions(const MetaInfo& info,
std::vector<bst_float>* out_preds,
const gbm::GBTreeModel& model) const {
size_t n = model.param.num_output_group * info.num_row;
const std::vector<bst_float>& base_margin = info.base_margin;
out_preds->resize(n);
if (base_margin.size() != 0) {
CHECK_EQ(out_preds->size(), n);
std::copy(base_margin.begin(), base_margin.end(), out_preds->begin());
} else {
std::fill(out_preds->begin(), out_preds->end(), model.base_margin);
}
}
Predictor* Predictor::Create(std::string name) {
auto* e = ::dmlc::Registry<PredictorReg>::Get()->Find(name);
if (e == nullptr) {
@@ -23,3 +57,13 @@ Predictor* Predictor::Create(std::string name) {
return (e->body)();
}
} // namespace xgboost
namespace xgboost {
namespace predictor {
// List of files that will be force linked in static links.
#ifdef XGBOOST_USE_CUDA
DMLC_REGISTRY_LINK_TAG(gpu_predictor);
#endif
DMLC_REGISTRY_LINK_TAG(cpu_predictor);
} // namespace predictor
} // namespace xgboost