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

Simplify inplace-predict. (#7910)

Browse Source 
Pass the `X` as part of Proxy DMatrix instead of an independent `dmlc::any`.
This commit is contained in:
Jiaming Yuan
2022-05-18 17:52:00 +08:00
committed by GitHub
parent 19775ffe15
commit 765097d514
17 changed files with 317 additions and 297 deletions
Download Patch File Download Diff File Expand all files Collapse all files

54
tests/cpp/predictor/test_predictor.cc
View File

@@ -2,19 +2,20 @@
* Copyright 2020-2021 by Contributors
*/
#include <gtest/gtest.h>
#include <xgboost/predictor.h>
#include <xgboost/data.h>
#include <xgboost/host_device_vector.h>
#include <xgboost/generic_parameters.h>
#include "test_predictor.h"
#include "../helpers.h"
#include "../../../src/data/adapter.h"
#include "../../../src/common/io.h"
#include "../../../src/common/categorical.h"
#include <gtest/gtest.h>
#include <xgboost/data.h>
#include <xgboost/generic_parameters.h>
#include <xgboost/host_device_vector.h>
#include <xgboost/predictor.h>
#include "../../../src/common/bitfield.h"
#include "../../../src/common/categorical.h"
#include "../../../src/common/io.h"
#include "../../../src/data/adapter.h"
#include "../../../src/data/proxy_dmatrix.h"
#include "../helpers.h"
namespace xgboost {
TEST(Predictor, PredictionCache) {
@@ -83,9 +84,8 @@ void TestTrainingPrediction(size_t rows, size_t bins,
train("gpu_predictor", &predictions_1);
}
void TestInplacePrediction(dmlc::any x, std::string predictor,
bst_row_t rows, bst_feature_t cols,
int32_t device) {
void TestInplacePrediction(std::shared_ptr<DMatrix> x, std::string predictor, bst_row_t rows,
bst_feature_t cols, int32_t device) {
size_t constexpr kClasses { 4 };
auto gen = RandomDataGenerator{rows, cols, 0.5}.Device(device);
std::shared_ptr<DMatrix> m = gen.GenerateDMatrix(true, false, kClasses);
@@ -105,24 +105,21 @@ void TestInplacePrediction(dmlc::any x, std::string predictor,
}
HostDeviceVector<float> *p_out_predictions_0{nullptr};
learner->InplacePredict(x, nullptr, PredictionType::kMargin,
std::numeric_limits<float>::quiet_NaN(),
learner->InplacePredict(x, PredictionType::kMargin, std::numeric_limits<float>::quiet_NaN(),
&p_out_predictions_0, 0, 2);
CHECK(p_out_predictions_0);
HostDeviceVector<float> predict_0 (p_out_predictions_0->Size());
predict_0.Copy(*p_out_predictions_0);
HostDeviceVector<float> *p_out_predictions_1{nullptr};
learner->InplacePredict(x, nullptr, PredictionType::kMargin,
std::numeric_limits<float>::quiet_NaN(),
learner->InplacePredict(x, PredictionType::kMargin, std::numeric_limits<float>::quiet_NaN(),
&p_out_predictions_1, 2, 4);
CHECK(p_out_predictions_1);
HostDeviceVector<float> predict_1 (p_out_predictions_1->Size());
predict_1.Copy(*p_out_predictions_1);
HostDeviceVector<float>* p_out_predictions{nullptr};
learner->InplacePredict(x, nullptr, PredictionType::kMargin,
std::numeric_limits<float>::quiet_NaN(),
learner->InplacePredict(x, PredictionType::kMargin, std::numeric_limits<float>::quiet_NaN(),
&p_out_predictions, 0, 4);
auto& h_pred = p_out_predictions->HostVector();
@@ -378,25 +375,28 @@ void TestSparsePrediction(float sparsity, std::string predictor) {
learner->SetParam("predictor", predictor);
learner->Predict(Xy, false, &sparse_predt, 0, 0);
std::vector<float> with_nan(kRows * kCols, std::numeric_limits<float>::quiet_NaN());
for (auto const& page : Xy->GetBatches<SparsePage>()) {
HostDeviceVector<float> with_nan(kRows * kCols, std::numeric_limits<float>::quiet_NaN());
auto& h_with_nan = with_nan.HostVector();
for (auto const &page : Xy->GetBatches<SparsePage>()) {
auto batch = page.GetView();
for (size_t i = 0; i < batch.Size(); ++i) {
auto row = batch[i];
for (auto e : row) {
with_nan[i * kCols + e.index] = e.fvalue;
h_with_nan[i * kCols + e.index] = e.fvalue;
}
}
}
learner->SetParam("predictor", "cpu_predictor");
// Xcode_12.4 doesn't compile with `std::make_shared`.
auto dense = std::shared_ptr<data::DenseAdapter>(
new data::DenseAdapter(with_nan.data(), kRows, kCols));
auto dense = std::shared_ptr<DMatrix>(new data::DMatrixProxy{});
auto array_interface = GetArrayInterface(&with_nan, kRows, kCols);
std::string arr_str;
Json::Dump(array_interface, &arr_str);
dynamic_cast<data::DMatrixProxy *>(dense.get())->SetArrayData(arr_str.data());
HostDeviceVector<float> *p_dense_predt;
learner->InplacePredict(dmlc::any(dense), nullptr, PredictionType::kValue,
std::numeric_limits<float>::quiet_NaN(), &p_dense_predt,
0, 0);
learner->InplacePredict(dense, PredictionType::kValue, std::numeric_limits<float>::quiet_NaN(),
&p_dense_predt, 0, 0);
auto const& dense_predt = *p_dense_predt;
if (predictor == "cpu_predictor") {