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

Prepare external memory support for hist. (#7638)

Browse Source 
This PR prepares the GHistIndexMatrix to host the column matrix which is used by the hist tree method by accepting sparse_threshold parameter.

Some cleanups are made to ensure the correct batch param is being passed into DMatrix along with some additional tests for correctness of SimpleDMatrix.
This commit is contained in:
Jiaming Yuan
2022-02-10 16:58:02 +08:00
committed by GitHub
parent 87c01f49d8
commit 2775c2a1ab
24 changed files with 368 additions and 201 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/cpp/data/test_ellpack_page.cu
View File

@@ -81,13 +81,13 @@ TEST(EllpackPage, BuildGidxSparse) {
TEST(EllpackPage, FromCategoricalBasic) {
using common::AsCat;
size_t constexpr kRows = 1000, kCats = 13, kCols = 1;
size_t max_bins = 8;
int32_t max_bins = 8;
auto x = GenerateRandomCategoricalSingleColumn(kRows, kCats);
auto m = GetDMatrixFromData(x, kRows, 1);
auto& h_ft = m->Info().feature_types.HostVector();
h_ft.resize(kCols, FeatureType::kCategorical);
BatchParam p(0, max_bins);
BatchParam p{0, max_bins};
auto ellpack = EllpackPage(m.get(), p);
auto accessor = ellpack.Impl()->GetDeviceAccessor(0);
ASSERT_EQ(kCats, accessor.NumBins());

22
tests/cpp/data/test_gradient_index.cc
View File

@@ -4,8 +4,8 @@
#include <gtest/gtest.h>
#include <xgboost/data.h>
#include "../helpers.h"
#include "../../../src/data/gradient_index.h"
#include "../helpers.h"
namespace xgboost {
namespace data {
@@ -13,12 +13,22 @@ TEST(GradientIndex, ExternalMemory) {
std::unique_ptr<DMatrix> dmat = CreateSparsePageDMatrix(10000);
std::vector<size_t> base_rowids;
std::vector<float> hessian(dmat->Info().num_row_, 1);
for (auto const &page : dmat->GetBatches<GHistIndexMatrix>(
{GenericParameter::kCpuId, 64, hessian})) {
for (auto const &page : dmat->GetBatches<GHistIndexMatrix>({64, hessian, true})) {
base_rowids.push_back(page.base_rowid);
}
size_t i = 0;
for (auto const& page : dmat->GetBatches<SparsePage>()) {
for (auto const &page : dmat->GetBatches<SparsePage>()) {
ASSERT_EQ(base_rowids[i], page.base_rowid);
++i;
}
base_rowids.clear();
for (auto const &page : dmat->GetBatches<GHistIndexMatrix>({64, hessian, false})) {
base_rowids.push_back(page.base_rowid);
}
i = 0;
for (auto const &page : dmat->GetBatches<SparsePage>()) {
ASSERT_EQ(base_rowids[i], page.base_rowid);
++i;
}
@@ -33,10 +43,10 @@ TEST(GradientIndex, FromCategoricalBasic) {
auto &h_ft = m->Info().feature_types.HostVector();
h_ft.resize(kCols, FeatureType::kCategorical);
BatchParam p(0, max_bins);
BatchParam p(max_bins, 0.8);
GHistIndexMatrix gidx;
gidx.Init(m.get(), max_bins, false, common::OmpGetNumThreads(0), {});
gidx.Init(m.get(), max_bins, p.sparse_thresh, false, common::OmpGetNumThreads(0), {});
auto x_copy = x;
std::sort(x_copy.begin(), x_copy.end());

4
tests/cpp/data/test_iterative_device_dmatrix.cu
View File

@@ -21,7 +21,7 @@ void TestEquivalent(float sparsity) {
std::unique_ptr<EllpackPageImpl> page_concatenated {
new EllpackPageImpl(0, first->Cuts(), first->is_dense,
first->row_stride, 1000 * 100)};
for (auto& batch : m.GetBatches<EllpackPage>()) {
for (auto& batch : m.GetBatches<EllpackPage>({})) {
auto page = batch.Impl();
size_t num_elements = page_concatenated->Copy(0, page, offset);
offset += num_elements;
@@ -93,7 +93,7 @@ TEST(IterativeDeviceDMatrix, RowMajor) {
0, 256);
size_t n_batches = 0;
std::string interface_str = iter.AsArray();
for (auto& ellpack : m.GetBatches<EllpackPage>()) {
for (auto& ellpack : m.GetBatches<EllpackPage>({})) {
n_batches ++;
auto impl = ellpack.Impl();
common::CompressedIterator<uint32_t> iterator(