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

merge 23Mar01

Browse Source
This commit is contained in:
amdsc21
2023-05-02 00:05:58 +02:00
parent 313a74b582 08ce495b5d
commit 5446c501af
258 changed files with 7471 additions and 5379 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
tests/cpp/tree/gpu_hist/test_gradient_based_sampler.cu
View File

@@ -1,12 +1,13 @@
/*!
* Copyright 2020-2021 by XGBoost Contributors
/**
* Copyright 2020-2023, XGBoost Contributors
*/
#include <gtest/gtest.h>
#include "../../../../src/data/ellpack_page.cuh"
#include "../../../../src/tree/gpu_hist/gradient_based_sampler.cuh"
#include "../../../../src/tree/param.h"
#include "../../filesystem.h" // dmlc::TemporaryDirectory
#include "../../../../src/tree/param.h" // TrainParam
#include "../../filesystem.h" // dmlc::TemporaryDirectory
#include "../../helpers.h"
namespace xgboost {
@@ -31,14 +32,15 @@ void VerifySampling(size_t page_size,
}
gpair.SetDevice(0);
BatchParam param{0, 256};
auto page = (*dmat->GetBatches<EllpackPage>(param).begin()).Impl();
Context ctx{MakeCUDACtx(0)};
auto param = BatchParam{256, tree::TrainParam::DftSparseThreshold()};
auto page = (*dmat->GetBatches<EllpackPage>(&ctx, param).begin()).Impl();
if (page_size != 0) {
EXPECT_NE(page->n_rows, kRows);
}
GradientBasedSampler sampler(page, kRows, param, subsample, sampling_method);
auto sample = sampler.Sample(gpair.DeviceSpan(), dmat.get());
GradientBasedSampler sampler(&ctx, page, kRows, param, subsample, sampling_method);
auto sample = sampler.Sample(&ctx, gpair.DeviceSpan(), dmat.get());
if (fixed_size_sampling) {
EXPECT_EQ(sample.sample_rows, kRows);
@@ -86,12 +88,13 @@ TEST(GradientBasedSampler, NoSamplingExternalMemory) {
auto gpair = GenerateRandomGradients(kRows);
gpair.SetDevice(0);
BatchParam param{0, 256};
auto page = (*dmat->GetBatches<EllpackPage>(param).begin()).Impl();
Context ctx{MakeCUDACtx(0)};
auto param = BatchParam{256, tree::TrainParam::DftSparseThreshold()};
auto page = (*dmat->GetBatches<EllpackPage>(&ctx, param).begin()).Impl();
EXPECT_NE(page->n_rows, kRows);
GradientBasedSampler sampler(page, kRows, param, kSubsample, TrainParam::kUniform);
auto sample = sampler.Sample(gpair.DeviceSpan(), dmat.get());
GradientBasedSampler sampler(&ctx, page, kRows, param, kSubsample, TrainParam::kUniform);
auto sample = sampler.Sample(&ctx, gpair.DeviceSpan(), dmat.get());
auto sampled_page = sample.page;
EXPECT_EQ(sample.sample_rows, kRows);
EXPECT_EQ(sample.gpair.size(), gpair.Size());
@@ -103,7 +106,7 @@ TEST(GradientBasedSampler, NoSamplingExternalMemory) {
ci(buffer.data(), sampled_page->NumSymbols());
size_t offset = 0;
for (auto& batch : dmat->GetBatches<EllpackPage>(param)) {
for (auto& batch : dmat->GetBatches<EllpackPage>(&ctx, param)) {
auto page = batch.Impl();
std::vector<common::CompressedByteT> page_buffer(page->gidx_buffer.HostVector());
common::CompressedIterator<common::CompressedByteT>

19
tests/cpp/tree/gpu_hist/test_histogram.cu
View File

@@ -1,9 +1,14 @@
/**
* Copyright 2020-2023, XGBoost Contributors
*/
#include <gtest/gtest.h>
#include <vector>
#include "../../../../src/common/categorical.h"
#include "../../../../src/tree/gpu_hist/histogram.cuh"
#include "../../../../src/tree/gpu_hist/row_partitioner.cuh"
#include "../../../../src/tree/param.h" // TrainParam
#include "../../categorical_helpers.h"
#include "../../helpers.h"
@@ -11,15 +16,15 @@ namespace xgboost {
namespace tree {
void TestDeterministicHistogram(bool is_dense, int shm_size) {
Context ctx = CreateEmptyGenericParam(0);
Context ctx = MakeCUDACtx(0);
size_t constexpr kBins = 256, kCols = 120, kRows = 16384, kRounds = 16;
float constexpr kLower = -1e-2, kUpper = 1e2;
float sparsity = is_dense ? 0.0f : 0.5f;
auto matrix = RandomDataGenerator(kRows, kCols, sparsity).GenerateDMatrix();
BatchParam batch_param{0, static_cast<int32_t>(kBins)};
auto batch_param = BatchParam{kBins, tree::TrainParam::DftSparseThreshold()};
for (auto const& batch : matrix->GetBatches<EllpackPage>(batch_param)) {
for (auto const& batch : matrix->GetBatches<EllpackPage>(&ctx, batch_param)) {
auto* page = batch.Impl();
tree::RowPartitioner row_partitioner(0, kRows);
@@ -132,13 +137,13 @@ void ValidateCategoricalHistogram(size_t n_categories, common::Span<GradientPair
// Test 1 vs rest categorical histogram is equivalent to one hot encoded data.
void TestGPUHistogramCategorical(size_t num_categories) {
auto ctx = CreateEmptyGenericParam(0);
auto ctx = MakeCUDACtx(0);
size_t constexpr kRows = 340;
size_t constexpr kBins = 256;
auto x = GenerateRandomCategoricalSingleColumn(kRows, num_categories);
auto cat_m = GetDMatrixFromData(x, kRows, 1);
cat_m->Info().feature_types.HostVector().push_back(FeatureType::kCategorical);
BatchParam batch_param{0, static_cast<int32_t>(kBins)};
auto batch_param = BatchParam{kBins, tree::TrainParam::DftSparseThreshold()};
tree::RowPartitioner row_partitioner(0, kRows);
auto ridx = row_partitioner.GetRows(0);
dh::device_vector<GradientPairInt64> cat_hist(num_categories);
@@ -148,7 +153,7 @@ void TestGPUHistogramCategorical(size_t num_categories) {
/**
* Generate hist with cat data.
*/
for (auto const &batch : cat_m->GetBatches<EllpackPage>(batch_param)) {
for (auto const &batch : cat_m->GetBatches<EllpackPage>(&ctx, batch_param)) {
auto* page = batch.Impl();
FeatureGroups single_group(page->Cuts());
BuildGradientHistogram(ctx.CUDACtx(), page->GetDeviceAccessor(0),
@@ -162,7 +167,7 @@ void TestGPUHistogramCategorical(size_t num_categories) {
auto x_encoded = OneHotEncodeFeature(x, num_categories);
auto encode_m = GetDMatrixFromData(x_encoded, kRows, num_categories);
dh::device_vector<GradientPairInt64> encode_hist(2 * num_categories);
for (auto const &batch : encode_m->GetBatches<EllpackPage>(batch_param)) {
for (auto const &batch : encode_m->GetBatches<EllpackPage>(&ctx, batch_param)) {
auto* page = batch.Impl();
FeatureGroups single_group(page->Cuts());
BuildGradientHistogram(ctx.CUDACtx(), page->GetDeviceAccessor(0),

4
tests/cpp/tree/hist/test_evaluate_splits.cc
View File

@@ -41,7 +41,7 @@ void TestEvaluateSplits(bool force_read_by_column) {
size_t constexpr kMaxBins = 4;
// dense, no missing values
GHistIndexMatrix gmat(dmat.get(), kMaxBins, 0.5, false, AllThreadsForTest());
GHistIndexMatrix gmat(&ctx, dmat.get(), kMaxBins, 0.5, false);
common::RowSetCollection row_set_collection;
std::vector<size_t> &row_indices = *row_set_collection.Data();
row_indices.resize(kRows);
@@ -228,7 +228,7 @@ auto CompareOneHotAndPartition(bool onehot) {
auto evaluator = HistEvaluator<CPUExpandEntry>{&ctx, &param, dmat->Info(), sampler};
std::vector<CPUExpandEntry> entries(1);
for (auto const &gmat : dmat->GetBatches<GHistIndexMatrix>({32, param.sparse_threshold})) {
for (auto const &gmat : dmat->GetBatches<GHistIndexMatrix>(&ctx, {32, param.sparse_threshold})) {
common::HistCollection hist;
entries.front().nid = 0;

76
tests/cpp/tree/hist/test_histogram.cc
View File

@@ -25,6 +25,7 @@ void InitRowPartitionForTest(common::RowSetCollection *row_set, size_t n_samples
} // anonymous namespace
void TestAddHistRows(bool is_distributed) {
auto ctx = CreateEmptyGenericParam(Context::kCpuId);
std::vector<CPUExpandEntry> nodes_for_explicit_hist_build_;
std::vector<CPUExpandEntry> nodes_for_subtraction_trick_;
int starting_index = std::numeric_limits<int>::max();
@@ -32,9 +33,9 @@ void TestAddHistRows(bool is_distributed) {
size_t constexpr kNRows = 8, kNCols = 16;
int32_t constexpr kMaxBins = 4;
auto p_fmat =
RandomDataGenerator(kNRows, kNCols, 0.8).Seed(3).GenerateDMatrix();
auto const &gmat = *(p_fmat->GetBatches<GHistIndexMatrix>(BatchParam{kMaxBins, 0.5}).begin());
auto p_fmat = RandomDataGenerator(kNRows, kNCols, 0.8).Seed(3).GenerateDMatrix();
auto const &gmat =
*(p_fmat->GetBatches<GHistIndexMatrix>(&ctx, BatchParam{kMaxBins, 0.5}).begin());
RegTree tree;
@@ -73,6 +74,7 @@ TEST(CPUHistogram, AddRows) {
void TestSyncHist(bool is_distributed) {
size_t constexpr kNRows = 8, kNCols = 16;
int32_t constexpr kMaxBins = 4;
auto ctx = CreateEmptyGenericParam(Context::kCpuId);
std::vector<CPUExpandEntry> nodes_for_explicit_hist_build_;
std::vector<CPUExpandEntry> nodes_for_subtraction_trick_;
@@ -80,9 +82,9 @@ void TestSyncHist(bool is_distributed) {
int sync_count = 0;
RegTree tree;
auto p_fmat =
RandomDataGenerator(kNRows, kNCols, 0.8).Seed(3).GenerateDMatrix();
auto const &gmat = *(p_fmat->GetBatches<GHistIndexMatrix>(BatchParam{kMaxBins, 0.5}).begin());
auto p_fmat = RandomDataGenerator(kNRows, kNCols, 0.8).Seed(3).GenerateDMatrix();
auto const &gmat =
*(p_fmat->GetBatches<GHistIndexMatrix>(&ctx, BatchParam{kMaxBins, 0.5}).begin());
HistogramBuilder<CPUExpandEntry> histogram;
uint32_t total_bins = gmat.cut.Ptrs().back();
@@ -227,12 +229,15 @@ TEST(CPUHistogram, SyncHist) {
void TestBuildHistogram(bool is_distributed, bool force_read_by_column, bool is_col_split) {
size_t constexpr kNRows = 8, kNCols = 16;
int32_t constexpr kMaxBins = 4;
auto p_fmat = RandomDataGenerator(kNRows, kNCols, 0.8).Seed(3).GenerateDMatrix();
auto ctx = CreateEmptyGenericParam(Context::kCpuId);
auto p_fmat =
RandomDataGenerator(kNRows, kNCols, 0.8).Seed(3).GenerateDMatrix();
if (is_col_split) {
p_fmat = std::shared_ptr<DMatrix>{
p_fmat->SliceCol(collective::GetWorldSize(), collective::GetRank())};
}
auto const &gmat = *(p_fmat->GetBatches<GHistIndexMatrix>(BatchParam{kMaxBins, 0.5}).begin());
auto const &gmat =
*(p_fmat->GetBatches<GHistIndexMatrix>(&ctx, BatchParam{kMaxBins, 0.5}).begin());
uint32_t total_bins = gmat.cut.Ptrs().back();
static double constexpr kEps = 1e-6;
@@ -257,9 +262,9 @@ void TestBuildHistogram(bool is_distributed, bool force_read_by_column, bool is_
CPUExpandEntry node{RegTree::kRoot, tree.GetDepth(0)};
std::vector<CPUExpandEntry> nodes_for_explicit_hist_build;
nodes_for_explicit_hist_build.push_back(node);
for (auto const &gidx : p_fmat->GetBatches<GHistIndexMatrix>({kMaxBins, 0.5})) {
histogram.BuildHist(0, gidx, &tree, row_set_collection,
nodes_for_explicit_hist_build, {}, gpair, force_read_by_column);
for (auto const &gidx : p_fmat->GetBatches<GHistIndexMatrix>(&ctx, {kMaxBins, 0.5})) {
histogram.BuildHist(0, gidx, &tree, row_set_collection, nodes_for_explicit_hist_build, {},
gpair, force_read_by_column);
}
// Check if number of histogram bins is correct
@@ -325,6 +330,8 @@ void TestHistogramCategorical(size_t n_categories, bool force_read_by_column) {
auto x = GenerateRandomCategoricalSingleColumn(kRows, n_categories);
auto cat_m = GetDMatrixFromData(x, kRows, 1);
cat_m->Info().feature_types.HostVector().push_back(FeatureType::kCategorical);
auto ctx = CreateEmptyGenericParam(Context::kCpuId);
BatchParam batch_param{0, static_cast<int32_t>(kBins)};
RegTree tree;
@@ -345,12 +352,11 @@ void TestHistogramCategorical(size_t n_categories, bool force_read_by_column) {
* Generate hist with cat data.
*/
HistogramBuilder<CPUExpandEntry> cat_hist;
for (auto const &gidx : cat_m->GetBatches<GHistIndexMatrix>({kBins, 0.5})) {
for (auto const &gidx : cat_m->GetBatches<GHistIndexMatrix>(&ctx, {kBins, 0.5})) {
auto total_bins = gidx.cut.TotalBins();
cat_hist.Reset(total_bins, {kBins, 0.5}, omp_get_max_threads(), 1, false, false);
cat_hist.BuildHist(0, gidx, &tree, row_set_collection,
nodes_for_explicit_hist_build, {}, gpair.HostVector(),
force_read_by_column);
cat_hist.BuildHist(0, gidx, &tree, row_set_collection, nodes_for_explicit_hist_build, {},
gpair.HostVector(), force_read_by_column);
}
/**
@@ -359,12 +365,11 @@ void TestHistogramCategorical(size_t n_categories, bool force_read_by_column) {
auto x_encoded = OneHotEncodeFeature(x, n_categories);
auto encode_m = GetDMatrixFromData(x_encoded, kRows, n_categories);
HistogramBuilder<CPUExpandEntry> onehot_hist;
for (auto const &gidx : encode_m->GetBatches<GHistIndexMatrix>({kBins, 0.5})) {
for (auto const &gidx : encode_m->GetBatches<GHistIndexMatrix>(&ctx, {kBins, 0.5})) {
auto total_bins = gidx.cut.TotalBins();
onehot_hist.Reset(total_bins, {kBins, 0.5}, omp_get_max_threads(), 1, false, false);
onehot_hist.BuildHist(0, gidx, &tree, row_set_collection, nodes_for_explicit_hist_build, {},
gpair.HostVector(),
force_read_by_column);
gpair.HostVector(), force_read_by_column);
}
auto cat = cat_hist.Histogram()[0];
@@ -382,8 +387,8 @@ TEST(CPUHistogram, Categorical) {
}
}
namespace {
void TestHistogramExternalMemory(BatchParam batch_param, bool is_approx, bool force_read_by_column) {
Context ctx;
void TestHistogramExternalMemory(Context const *ctx, BatchParam batch_param, bool is_approx,
bool force_read_by_column) {
size_t constexpr kEntries = 1 << 16;
auto m = CreateSparsePageDMatrix(kEntries, "cache");
@@ -410,7 +415,7 @@ void TestHistogramExternalMemory(BatchParam batch_param, bool is_approx, bool fo
* Multi page
*/
std::vector<common::RowSetCollection> rows_set;
for (auto const &page : m->GetBatches<GHistIndexMatrix>(batch_param)) {
for (auto const &page : m->GetBatches<GHistIndexMatrix>(ctx, batch_param)) {
CHECK_LT(page.base_rowid, m->Info().num_row_);
auto n_rows_in_node = page.Size();
partition_size[0] = std::max(partition_size[0], n_rows_in_node);
@@ -426,12 +431,12 @@ void TestHistogramExternalMemory(BatchParam batch_param, bool is_approx, bool fo
1, [&](size_t nidx_in_set) { return partition_size.at(nidx_in_set); },
256};
multi_build.Reset(total_bins, batch_param, ctx.Threads(), rows_set.size(), false, false);
multi_build.Reset(total_bins, batch_param, ctx->Threads(), rows_set.size(), false, false);
size_t page_idx{0};
for (auto const &page : m->GetBatches<GHistIndexMatrix>(batch_param)) {
multi_build.BuildHist(page_idx, space, page, &tree, rows_set.at(page_idx), nodes, {},
h_gpair, force_read_by_column);
for (auto const &page : m->GetBatches<GHistIndexMatrix>(ctx, batch_param)) {
multi_build.BuildHist(page_idx, space, page, &tree, rows_set.at(page_idx), nodes, {}, h_gpair,
force_read_by_column);
++page_idx;
}
ASSERT_EQ(page_idx, 2);
@@ -447,16 +452,16 @@ void TestHistogramExternalMemory(BatchParam batch_param, bool is_approx, bool fo
common::RowSetCollection row_set_collection;
InitRowPartitionForTest(&row_set_collection, n_samples);
single_build.Reset(total_bins, batch_param, ctx.Threads(), 1, false, false);
single_build.Reset(total_bins, batch_param, ctx->Threads(), 1, false, false);
SparsePage concat;
std::vector<float> hess(m->Info().num_row_, 1.0f);
for (auto const& page : m->GetBatches<SparsePage>()) {
concat.Push(page);
}
auto cut = common::SketchOnDMatrix(m.get(), batch_param.max_bin, ctx.Threads(), false, hess);
auto cut = common::SketchOnDMatrix(ctx, m.get(), batch_param.max_bin, false, hess);
GHistIndexMatrix gmat(concat, {}, cut, batch_param.max_bin, false,
std::numeric_limits<double>::quiet_NaN(), ctx.Threads());
std::numeric_limits<double>::quiet_NaN(), ctx->Threads());
single_build.BuildHist(0, gmat, &tree, row_set_collection, nodes, {}, h_gpair, force_read_by_column);
single_page = single_build.Histogram()[0];
}
@@ -470,16 +475,17 @@ void TestHistogramExternalMemory(BatchParam batch_param, bool is_approx, bool fo
TEST(CPUHistogram, ExternalMemory) {
int32_t constexpr kBins = 256;
TestHistogramExternalMemory(BatchParam{kBins, common::Span<float>{}, false}, true, false);
TestHistogramExternalMemory(BatchParam{kBins, common::Span<float>{}, false}, true, true);
auto ctx = CreateEmptyGenericParam(Context::kCpuId);
TestHistogramExternalMemory(&ctx, BatchParam{kBins, common::Span<float>{}, false}, true, false);
TestHistogramExternalMemory(&ctx, BatchParam{kBins, common::Span<float>{}, false}, true, true);
float sparse_thresh{0.5};
TestHistogramExternalMemory({kBins, sparse_thresh}, false, false);
TestHistogramExternalMemory({kBins, sparse_thresh}, false, true);
TestHistogramExternalMemory(&ctx, {kBins, sparse_thresh}, false, false);
TestHistogramExternalMemory(&ctx, {kBins, sparse_thresh}, false, true);
sparse_thresh = std::numeric_limits<float>::quiet_NaN();
TestHistogramExternalMemory({kBins, sparse_thresh}, false, false);
TestHistogramExternalMemory({kBins, sparse_thresh}, false, true);
TestHistogramExternalMemory(&ctx, {kBins, sparse_thresh}, false, false);
TestHistogramExternalMemory(&ctx, {kBins, sparse_thresh}, false, true);
}
} // namespace tree
} // namespace xgboost

6
tests/cpp/tree/test_approx.cc
View File

@@ -34,7 +34,7 @@ TEST(Approx, Partitioner) {
std::vector<CPUExpandEntry> candidates{{0, 0}};
candidates.front().split.loss_chg = 0.4;
for (auto const& page : Xy->GetBatches<GHistIndexMatrix>({64, hess, true})) {
for (auto const& page : Xy->GetBatches<GHistIndexMatrix>(&ctx, {64, hess, true})) {
bst_feature_t const split_ind = 0;
{
auto min_value = page.cut.MinValues()[split_ind];
@@ -84,7 +84,7 @@ void TestColumnSplitPartitioner(size_t n_samples, size_t base_rowid, std::shared
Context ctx;
ctx.InitAllowUnknown(Args{});
for (auto const& page : dmat->GetBatches<GHistIndexMatrix>({64, *hess, true})) {
for (auto const& page : dmat->GetBatches<GHistIndexMatrix>(&ctx, {64, *hess, true})) {
{
RegTree tree;
CommonRowPartitioner partitioner{&ctx, n_samples, base_rowid, true};
@@ -133,7 +133,7 @@ TEST(Approx, PartitionerColSplit) {
Context ctx;
ctx.InitAllowUnknown(Args{});
CommonRowPartitioner mid_partitioner{&ctx, n_samples, base_rowid, false};
for (auto const& page : Xy->GetBatches<GHistIndexMatrix>({64, hess, true})) {
for (auto const& page : Xy->GetBatches<GHistIndexMatrix>(&ctx, {64, hess, true})) {
bst_feature_t const split_ind = 0;
min_value = page.cut.MinValues()[split_ind];

2
tests/cpp/tree/test_common_partitioner.cc
View File

@@ -43,7 +43,7 @@ void TestLeafPartition(size_t n_samples) {
std::vector<size_t> h_nptr;
float split_value{0};
for (auto const& page : Xy->GetBatches<GHistIndexMatrix>({Context::kCpuId, 64})) {
for (auto const& page : Xy->GetBatches<GHistIndexMatrix>(&ctx, BatchParam{64, 0.2})) {
bst_feature_t const split_ind = 0;
auto ptr = page.cut.Ptrs()[split_ind + 1];
split_value = page.cut.Values().at(ptr / 2);

41
tests/cpp/tree/test_gpu_hist.cu
View File

@@ -218,17 +218,16 @@ TEST(GpuHist, TestHistogramIndex) {
TestHistogramIndexImpl();
}
void UpdateTree(HostDeviceVector<GradientPair>* gpair, DMatrix* dmat,
size_t gpu_page_size, RegTree* tree,
HostDeviceVector<bst_float>* preds, float subsample = 1.0f,
const std::string& sampling_method = "uniform",
void UpdateTree(Context const* ctx, HostDeviceVector<GradientPair>* gpair, DMatrix* dmat,
size_t gpu_page_size, RegTree* tree, HostDeviceVector<bst_float>* preds,
float subsample = 1.0f, const std::string& sampling_method = "uniform",
int max_bin = 2) {
if (gpu_page_size > 0) {
// Loop over the batches and count the records
int64_t batch_count = 0;
int64_t row_count = 0;
for (const auto& batch : dmat->GetBatches<EllpackPage>({0, max_bin})) {
for (const auto& batch : dmat->GetBatches<EllpackPage>(
ctx, BatchParam{max_bin, TrainParam::DftSparseThreshold()})) {
EXPECT_LT(batch.Size(), dmat->Info().num_row_);
batch_count++;
row_count += batch.Size();
@@ -249,14 +248,13 @@ void UpdateTree(HostDeviceVector<GradientPair>* gpair, DMatrix* dmat,
TrainParam param;
param.UpdateAllowUnknown(args);
Context ctx(CreateEmptyGenericParam(0));
ObjInfo task{ObjInfo::kRegression};
tree::GPUHistMaker hist_maker{&ctx, &task};
tree::GPUHistMaker hist_maker{ctx, &task};
std::vector<HostDeviceVector<bst_node_t>> position(1);
hist_maker.Update(&param, gpair, dmat, common::Span<HostDeviceVector<bst_node_t>>{position},
{tree});
auto cache = linalg::MakeTensorView(&ctx, preds->DeviceSpan(), preds->Size(), 1);
auto cache = linalg::MakeTensorView(ctx, preds->DeviceSpan(), preds->Size(), 1);
hist_maker.UpdatePredictionCache(dmat, cache);
}
@@ -274,12 +272,13 @@ TEST(GpuHist, UniformSampling) {
// Build a tree using the in-memory DMatrix.
RegTree tree;
HostDeviceVector<bst_float> preds(kRows, 0.0, 0);
UpdateTree(&gpair, dmat.get(), 0, &tree, &preds, 1.0, "uniform", kRows);
Context ctx(CreateEmptyGenericParam(0));
UpdateTree(&ctx, &gpair, dmat.get(), 0, &tree, &preds, 1.0, "uniform", kRows);
// Build another tree using sampling.
RegTree tree_sampling;
HostDeviceVector<bst_float> preds_sampling(kRows, 0.0, 0);
UpdateTree(&gpair, dmat.get(), 0, &tree_sampling, &preds_sampling, kSubsample,
"uniform", kRows);
UpdateTree(&ctx, &gpair, dmat.get(), 0, &tree_sampling, &preds_sampling, kSubsample, "uniform",
kRows);
// Make sure the predictions are the same.
auto preds_h = preds.ConstHostVector();
@@ -303,12 +302,13 @@ TEST(GpuHist, GradientBasedSampling) {
// Build a tree using the in-memory DMatrix.
RegTree tree;
HostDeviceVector<bst_float> preds(kRows, 0.0, 0);
UpdateTree(&gpair, dmat.get(), 0, &tree, &preds, 1.0, "uniform", kRows);
Context ctx(CreateEmptyGenericParam(0));
UpdateTree(&ctx, &gpair, dmat.get(), 0, &tree, &preds, 1.0, "uniform", kRows);
// Build another tree using sampling.
RegTree tree_sampling;
HostDeviceVector<bst_float> preds_sampling(kRows, 0.0, 0);
UpdateTree(&gpair, dmat.get(), 0, &tree_sampling, &preds_sampling, kSubsample,
UpdateTree(&ctx, &gpair, dmat.get(), 0, &tree_sampling, &preds_sampling, kSubsample,
"gradient_based", kRows);
// Make sure the predictions are the same.
@@ -337,12 +337,13 @@ TEST(GpuHist, ExternalMemory) {
// Build a tree using the in-memory DMatrix.
RegTree tree;
Context ctx(CreateEmptyGenericParam(0));
HostDeviceVector<bst_float> preds(kRows, 0.0, 0);
UpdateTree(&gpair, dmat.get(), 0, &tree, &preds, 1.0, "uniform", kRows);
UpdateTree(&ctx, &gpair, dmat.get(), 0, &tree, &preds, 1.0, "uniform", kRows);
// Build another tree using multiple ELLPACK pages.
RegTree tree_ext;
HostDeviceVector<bst_float> preds_ext(kRows, 0.0, 0);
UpdateTree(&gpair, dmat_ext.get(), kPageSize, &tree_ext, &preds_ext, 1.0, "uniform", kRows);
UpdateTree(&ctx, &gpair, dmat_ext.get(), kPageSize, &tree_ext, &preds_ext, 1.0, "uniform", kRows);
// Make sure the predictions are the same.
auto preds_h = preds.ConstHostVector();
@@ -374,17 +375,17 @@ TEST(GpuHist, ExternalMemoryWithSampling) {
// Build a tree using the in-memory DMatrix.
auto rng = common::GlobalRandom();
Context ctx(CreateEmptyGenericParam(0));
RegTree tree;
HostDeviceVector<bst_float> preds(kRows, 0.0, 0);
UpdateTree(&gpair, dmat.get(), 0, &tree, &preds, kSubsample, kSamplingMethod,
kRows);
UpdateTree(&ctx, &gpair, dmat.get(), 0, &tree, &preds, kSubsample, kSamplingMethod, kRows);
// Build another tree using multiple ELLPACK pages.
common::GlobalRandom() = rng;
RegTree tree_ext;
HostDeviceVector<bst_float> preds_ext(kRows, 0.0, 0);
UpdateTree(&gpair, dmat_ext.get(), kPageSize, &tree_ext, &preds_ext,
kSubsample, kSamplingMethod, kRows);
UpdateTree(&ctx, &gpair, dmat_ext.get(), kPageSize, &tree_ext, &preds_ext, kSubsample,
kSamplingMethod, kRows);
// Make sure the predictions are the same.
auto preds_h = preds.ConstHostVector();

2
tests/cpp/tree/test_quantile_hist.cc
View File

@@ -36,7 +36,7 @@ void TestPartitioner(bst_target_t n_targets) {
std::vector<ExpandEntry> candidates{{0, 0}};
candidates.front().split.loss_chg = 0.4;
auto cuts = common::SketchOnDMatrix(Xy.get(), 64, ctx.Threads());
auto cuts = common::SketchOnDMatrix(&ctx, Xy.get(), 64);
for (auto const& page : Xy->GetBatches<SparsePage>()) {
GHistIndexMatrix gmat(page, {}, cuts, 64, true, 0.5, ctx.Threads());

15
tests/cpp/tree/test_regen.cc
View File

@@ -15,16 +15,17 @@ class DMatrixForTest : public data::SimpleDMatrix {
public:
using SimpleDMatrix::SimpleDMatrix;
BatchSet<GHistIndexMatrix> GetGradientIndex(const BatchParam& param) override {
BatchSet<GHistIndexMatrix> GetGradientIndex(Context const* ctx,
const BatchParam& param) override {
auto backup = this->gradient_index_;
auto iter = SimpleDMatrix::GetGradientIndex(param);
auto iter = SimpleDMatrix::GetGradientIndex(ctx, param);
n_regen_ += (backup != this->gradient_index_);
return iter;
}
BatchSet<EllpackPage> GetEllpackBatches(const BatchParam& param) override {
BatchSet<EllpackPage> GetEllpackBatches(Context const* ctx, const BatchParam& param) override {
auto backup = this->ellpack_page_;
auto iter = SimpleDMatrix::GetEllpackBatches(param);
auto iter = SimpleDMatrix::GetEllpackBatches(ctx, param);
n_regen_ += (backup != this->ellpack_page_);
return iter;
}
@@ -50,8 +51,8 @@ class RegenTest : public ::testing::Test {
HostDeviceVector<float> storage;
auto dense = RandomDataGenerator{kRows, kCols, 0.5}.GenerateArrayInterface(&storage);
auto adapter = data::ArrayAdapter(StringView{dense});
p_fmat_ = std::shared_ptr<DMatrix>(new DMatrixForTest{
&adapter, std::numeric_limits<float>::quiet_NaN(), AllThreadsForTest()});
p_fmat_ = std::shared_ptr<DMatrix>(
new DMatrixForTest{&adapter, std::numeric_limits<float>::quiet_NaN(), AllThreadsForTest()});
p_fmat_->Info().labels.Reshape(256, 1);
auto labels = p_fmat_->Info().labels.Data();
@@ -74,7 +75,7 @@ class RegenTest : public ::testing::Test {
auto for_test = dynamic_cast<DMatrixForTest*>(p_fmat_.get());
CHECK(for_test);
auto backup = for_test->NumRegen();
for_test->GetBatches<Page>(BatchParam{});
for_test->GetBatches<Page>(p_fmat_->Ctx(), BatchParam{});
CHECK_EQ(for_test->NumRegen(), backup);
if (reset) {