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

sync up May15 2023

Browse Source
This commit is contained in:
amdsc21
2023-05-15 18:59:18 +02:00
parent b066accad6 7375bd058b
commit 8cad8c693c
37 changed files with 628 additions and 398 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
tests/cpp/tree/test_constraints.cc
View File

@@ -6,6 +6,8 @@
#include <string>
#include "../../../src/tree/constraints.h"
#include "../../../src/tree/hist/evaluate_splits.h"
#include "../helpers.h"
namespace xgboost {
namespace tree {
@@ -56,5 +58,37 @@ TEST(CPUFeatureInteractionConstraint, Basic) {
ASSERT_FALSE(constraints.Query(1, 5));
}
TEST(CPUMonoConstraint, Basic) {
std::size_t kRows{64}, kCols{16};
Context ctx;
TrainParam param;
std::vector<std::int32_t> mono(kCols, 1);
I32Array arr;
for (std::size_t i = 0; i < kCols; ++i) {
arr.GetArray().push_back(mono[i]);
}
Json jarr{std::move(arr)};
std::string str_mono;
Json::Dump(jarr, &str_mono);
str_mono.front() = '(';
str_mono.back() = ')';
param.UpdateAllowUnknown(Args{{"monotone_constraints", str_mono}});
auto Xy = RandomDataGenerator{kRows, kCols, 0.0}.GenerateDMatrix(true);
auto sampler = std::make_shared<common::ColumnSampler>();
HistEvaluator<CPUExpandEntry> evalutor{&ctx, &param, Xy->Info(), sampler};
evalutor.InitRoot(GradStats{2.0, 2.0});
SplitEntry split;
split.Update(1.0f, 0, 3.0, false, false, GradStats{1.0, 1.0}, GradStats{1.0, 1.0});
CPUExpandEntry entry{0, 0, split};
RegTree tree{1, static_cast<bst_feature_t>(kCols)};
evalutor.ApplyTreeSplit(entry, &tree);
ASSERT_TRUE(evalutor.Evaluator().has_constraint);
}
} // namespace tree
} // namespace xgboost

13
tests/cpp/tree/test_histmaker.cc
View File

@@ -90,13 +90,16 @@ void TestColumnSplit(int32_t rows, bst_feature_t cols, RegTree const& expected_t
param.Init(Args{});
updater->Update(&param, p_gradients.get(), sliced.get(), position, {&tree});
EXPECT_EQ(tree.NumExtraNodes(), 10);
EXPECT_EQ(tree[0].SplitIndex(), 1);
ASSERT_EQ(tree.NumExtraNodes(), 10);
ASSERT_EQ(tree[0].SplitIndex(), 1);
EXPECT_NE(tree[tree[0].LeftChild()].SplitIndex(), 0);
EXPECT_NE(tree[tree[0].RightChild()].SplitIndex(), 0);
ASSERT_NE(tree[tree[0].LeftChild()].SplitIndex(), 0);
ASSERT_NE(tree[tree[0].RightChild()].SplitIndex(), 0);
EXPECT_EQ(tree, expected_tree);
FeatureMap fmap;
auto json = tree.DumpModel(fmap, false, "json");
auto expected_json = expected_tree.DumpModel(fmap, false, "json");
ASSERT_EQ(json, expected_json);
}
} // anonymous namespace

111
tests/cpp/tree/test_quantile_hist.cc
View File

@@ -19,6 +19,8 @@
#include "xgboost/data.h"
namespace xgboost::tree {
namespace {
template <typename ExpandEntry>
void TestPartitioner(bst_target_t n_targets) {
std::size_t n_samples = 1024, base_rowid = 0;
@@ -86,8 +88,117 @@ void TestPartitioner(bst_target_t n_targets) {
}
}
}
} // anonymous namespace
TEST(QuantileHist, Partitioner) { TestPartitioner<CPUExpandEntry>(1); }
TEST(QuantileHist, MultiPartitioner) { TestPartitioner<MultiExpandEntry>(3); }
namespace {
template <typename ExpandEntry>
void VerifyColumnSplitPartitioner(bst_target_t n_targets, size_t n_samples,
bst_feature_t n_features, size_t base_rowid,
std::shared_ptr<DMatrix> Xy, float min_value, float mid_value,
CommonRowPartitioner const& expected_mid_partitioner) {
auto dmat =
std::unique_ptr<DMatrix>{Xy->SliceCol(collective::GetWorldSize(), collective::GetRank())};
Context ctx;
ctx.InitAllowUnknown(Args{});
std::vector<ExpandEntry> candidates{{0, 0}};
candidates.front().split.loss_chg = 0.4;
auto cuts = common::SketchOnDMatrix(&ctx, dmat.get(), 64);
for (auto const& page : Xy->GetBatches<SparsePage>()) {
GHistIndexMatrix gmat(page, {}, cuts, 64, true, 0.5, ctx.Threads());
bst_feature_t const split_ind = 0;
common::ColumnMatrix column_indices;
column_indices.InitFromSparse(page, gmat, 0.5, ctx.Threads());
{
RegTree tree{n_targets, n_features};
CommonRowPartitioner partitioner{&ctx, n_samples, base_rowid, true};
if constexpr (std::is_same<ExpandEntry, CPUExpandEntry>::value) {
GetSplit(&tree, min_value, &candidates);
} else {
GetMultiSplitForTest(&tree, min_value, &candidates);
}
partitioner.UpdatePosition<false, true>(&ctx, gmat, column_indices, candidates, &tree);
ASSERT_EQ(partitioner.Size(), 3);
ASSERT_EQ(partitioner[1].Size(), 0);
ASSERT_EQ(partitioner[2].Size(), n_samples);
}
{
RegTree tree{n_targets, n_features};
CommonRowPartitioner partitioner{&ctx, n_samples, base_rowid, true};
if constexpr (std::is_same<ExpandEntry, CPUExpandEntry>::value) {
GetSplit(&tree, mid_value, &candidates);
} else {
GetMultiSplitForTest(&tree, mid_value, &candidates);
}
auto left_nidx = tree.LeftChild(RegTree::kRoot);
partitioner.UpdatePosition<false, true>(&ctx, gmat, column_indices, candidates, &tree);
auto elem = partitioner[left_nidx];
ASSERT_LT(elem.Size(), n_samples);
ASSERT_GT(elem.Size(), 1);
auto expected_elem = expected_mid_partitioner[left_nidx];
ASSERT_EQ(elem.Size(), expected_elem.Size());
for (auto it = elem.begin, eit = expected_elem.begin; it != elem.end; ++it, ++eit) {
ASSERT_EQ(*it, *eit);
}
auto right_nidx = tree.RightChild(RegTree::kRoot);
elem = partitioner[right_nidx];
expected_elem = expected_mid_partitioner[right_nidx];
ASSERT_EQ(elem.Size(), expected_elem.Size());
for (auto it = elem.begin, eit = expected_elem.begin; it != elem.end; ++it, ++eit) {
ASSERT_EQ(*it, *eit);
}
}
}
}
template <typename ExpandEntry>
void TestColumnSplitPartitioner(bst_target_t n_targets) {
std::size_t n_samples = 1024, base_rowid = 0;
bst_feature_t n_features = 16;
auto Xy = RandomDataGenerator{n_samples, n_features, 0}.GenerateDMatrix(true);
std::vector<ExpandEntry> candidates{{0, 0}};
candidates.front().split.loss_chg = 0.4;
Context ctx;
ctx.InitAllowUnknown(Args{});
auto cuts = common::SketchOnDMatrix(&ctx, Xy.get(), 64);
float min_value, mid_value;
CommonRowPartitioner mid_partitioner{&ctx, n_samples, base_rowid, false};
for (auto const& page : Xy->GetBatches<SparsePage>()) {
GHistIndexMatrix gmat(page, {}, cuts, 64, true, 0.5, ctx.Threads());
bst_feature_t const split_ind = 0;
common::ColumnMatrix column_indices;
column_indices.InitFromSparse(page, gmat, 0.5, ctx.Threads());
min_value = gmat.cut.MinValues()[split_ind];
auto ptr = gmat.cut.Ptrs()[split_ind + 1];
mid_value = gmat.cut.Values().at(ptr / 2);
RegTree tree{n_targets, n_features};
if constexpr (std::is_same<ExpandEntry, CPUExpandEntry>::value) {
GetSplit(&tree, mid_value, &candidates);
} else {
GetMultiSplitForTest(&tree, mid_value, &candidates);
}
mid_partitioner.UpdatePosition<false, true>(&ctx, gmat, column_indices, candidates, &tree);
}
auto constexpr kWorkers = 4;
RunWithInMemoryCommunicator(kWorkers, VerifyColumnSplitPartitioner<ExpandEntry>, n_targets,
n_samples, n_features, base_rowid, Xy, min_value, mid_value, mid_partitioner);
}
} // anonymous namespace
TEST(QuantileHist, PartitionerColSplit) { TestColumnSplitPartitioner<CPUExpandEntry>(1); }
TEST(QuantileHist, MultiPartitionerColSplit) { TestColumnSplitPartitioner<MultiExpandEntry>(3); }
} // namespace xgboost::tree