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

Pass DMatrix into metric for caching. (#8790)

Browse Source
This commit is contained in:
Jiaming Yuan
2023-02-13 22:15:05 +08:00
committed by GitHub
parent 31d3ec07af
commit 81b2ee1153
17 changed files with 95 additions and 70 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tests/cpp/helpers.cc
View File

@@ -156,14 +156,15 @@ double GetMultiMetricEval(xgboost::Metric* metric,
xgboost::linalg::Tensor<float, 2> const& labels,
std::vector<xgboost::bst_float> weights,
std::vector<xgboost::bst_uint> groups) {
xgboost::MetaInfo info;
std::shared_ptr<xgboost::DMatrix> p_fmat{xgboost::RandomDataGenerator{0, 0, 0}.GenerateDMatrix()};
auto& info = p_fmat->Info();
info.num_row_ = labels.Shape(0);
info.labels.Reshape(labels.Shape()[0], labels.Shape()[1]);
info.labels.Data()->Copy(*labels.Data());
info.weights_.HostVector() = weights;
info.group_ptr_ = groups;
return metric->Eval(preds, info);
return metric->Evaluate(preds, p_fmat);
}
namespace xgboost {
@@ -661,4 +662,4 @@ void DeleteRMMResource(RMMAllocator*) {}
RMMAllocatorPtr SetUpRMMResourceForCppTests(int, char**) { return {nullptr, DeleteRMMResource}; }
#endif // !defined(XGBOOST_USE_RMM) || XGBOOST_USE_RMM != 1
} // namespace xgboost
} // namespace xgboost

5
tests/cpp/helpers.h
View File

@@ -301,6 +301,11 @@ class RandomDataGenerator {
std::shared_ptr<DMatrix> GenerateQuantileDMatrix();
};
// Generate an empty DMatrix, mostly for its meta info.
inline std::shared_ptr<DMatrix> EmptyDMatrix() {
return RandomDataGenerator{0, 0, 0.0}.GenerateDMatrix();
}
inline std::vector<float>
GenerateRandomCategoricalSingleColumn(int n, size_t num_categories) {
std::vector<float> x(n);

7
tests/cpp/metric/test_auc.cc
View File

@@ -20,12 +20,13 @@ TEST(Metric, DeclareUnifiedTest(BinaryAUC)) {
EXPECT_NEAR(GetMetricEval(metric, {1, 0, 0}, {0, 0, 1}), 0.25f, 1e-10);
// Invalid dataset
MetaInfo info;
auto p_fmat = EmptyDMatrix();
MetaInfo& info = p_fmat->Info();
info.labels = linalg::Tensor<float, 2>{{0.0f, 0.0f}, {2}, -1};
float auc = metric->Eval({1, 1}, info);
float auc = metric->Evaluate({1, 1}, p_fmat);
ASSERT_TRUE(std::isnan(auc));
*info.labels.Data() = HostDeviceVector<float>{};
auc = metric->Eval(HostDeviceVector<float>{}, info);
auc = metric->Evaluate(HostDeviceVector<float>{}, p_fmat);
ASSERT_TRUE(std::isnan(auc));
EXPECT_NEAR(GetMetricEval(metric, {0, 1, 0, 1}, {0, 1, 0, 1}), 1.0f, 1e-10);

7
tests/cpp/metric/test_elementwise_metric.cc
View File

@@ -19,7 +19,8 @@ inline void CheckDeterministicMetricElementWise(StringView name, int32_t device)
HostDeviceVector<float> predts;
size_t n_samples = 2048;
MetaInfo info;
auto p_fmat = EmptyDMatrix();
MetaInfo& info = p_fmat->Info();
info.labels.Reshape(n_samples, 1);
info.num_row_ = n_samples;
auto &h_labels = info.labels.Data()->HostVector();
@@ -36,9 +37,9 @@ inline void CheckDeterministicMetricElementWise(StringView name, int32_t device)
h_labels[i] = dist(&lcg);
}
auto result = metric->Eval(predts, info);
auto result = metric->Evaluate(predts, p_fmat);
for (size_t i = 0; i < 8; ++i) {
ASSERT_EQ(metric->Eval(predts, info), result);
ASSERT_EQ(metric->Evaluate(predts, p_fmat), result);
}
}
} // anonymous namespace

7
tests/cpp/metric/test_multiclass_metric.cc
View File

@@ -10,7 +10,8 @@ inline void CheckDeterministicMetricMultiClass(StringView name, int32_t device)
std::unique_ptr<Metric> metric{Metric::Create(name.c_str(), &ctx)};
HostDeviceVector<float> predts;
MetaInfo info;
auto p_fmat = EmptyDMatrix();
MetaInfo& info = p_fmat->Info();
auto &h_predts = predts.HostVector();
SimpleLCG lcg;
@@ -35,9 +36,9 @@ inline void CheckDeterministicMetricMultiClass(StringView name, int32_t device)
}
}
auto result = metric->Eval(predts, info);
auto result = metric->Evaluate(predts, p_fmat);
for (size_t i = 0; i < 8; ++i) {
ASSERT_EQ(metric->Eval(predts, info), result);
ASSERT_EQ(metric->Evaluate(predts, p_fmat), result);
}
}
} // namespace xgboost

25
tests/cpp/metric/test_survival_metric.cu
View File

@@ -18,7 +18,8 @@ inline void CheckDeterministicMetricElementWise(StringView name, int32_t device)
metric->Configure(Args{});
HostDeviceVector<float> predts;
MetaInfo info;
auto p_fmat = EmptyDMatrix();
MetaInfo& info = p_fmat->Info();
auto &h_predts = predts.HostVector();
SimpleLCG lcg;
@@ -40,9 +41,9 @@ inline void CheckDeterministicMetricElementWise(StringView name, int32_t device)
h_upper[i] = 10;
}
auto result = metric->Eval(predts, info);
auto result = metric->Evaluate(predts, p_fmat);
for (size_t i = 0; i < 8; ++i) {
ASSERT_EQ(metric->Eval(predts, info), result);
ASSERT_EQ(metric->Evaluate(predts, p_fmat), result);
}
}
} // anonymous namespace
@@ -54,7 +55,8 @@ TEST(Metric, DeclareUnifiedTest(AFTNegLogLik)) {
* Test aggregate output from the AFT metric over a small test data set.
* This is unlike AFTLoss.* tests, which verify metric values over individual data points.
**/
MetaInfo info;
auto p_fmat = EmptyDMatrix();
MetaInfo& info = p_fmat->Info();
info.num_row_ = 4;
info.labels_lower_bound_.HostVector()
= { 100.0f, 0.0f, 60.0f, 16.0f };
@@ -72,14 +74,15 @@ TEST(Metric, DeclareUnifiedTest(AFTNegLogLik)) {
std::unique_ptr<Metric> metric(Metric::Create("aft-nloglik", &ctx));
metric->Configure({ {"aft_loss_distribution", test_case.dist_type},
{"aft_loss_distribution_scale", "1.0"} });
EXPECT_NEAR(metric->Eval(preds, info), test_case.reference_value, 1e-4);
EXPECT_NEAR(metric->Evaluate(preds, p_fmat), test_case.reference_value, 1e-4);
}
}
TEST(Metric, DeclareUnifiedTest(IntervalRegressionAccuracy)) {
auto ctx = xgboost::CreateEmptyGenericParam(GPUIDX);
MetaInfo info;
auto p_fmat = EmptyDMatrix();
MetaInfo& info = p_fmat->Info();
info.num_row_ = 4;
info.labels_lower_bound_.HostVector() = { 20.0f, 0.0f, 60.0f, 16.0f };
info.labels_upper_bound_.HostVector() = { 80.0f, 20.0f, 80.0f, 200.0f };
@@ -87,15 +90,15 @@ TEST(Metric, DeclareUnifiedTest(IntervalRegressionAccuracy)) {
HostDeviceVector<bst_float> preds(4, std::log(60.0f));
std::unique_ptr<Metric> metric(Metric::Create("interval-regression-accuracy", &ctx));
EXPECT_FLOAT_EQ(metric->Eval(preds, info), 0.75f);
EXPECT_FLOAT_EQ(metric->Evaluate(preds, p_fmat), 0.75f);
info.labels_lower_bound_.HostVector()[2] = 70.0f;
EXPECT_FLOAT_EQ(metric->Eval(preds, info), 0.50f);
EXPECT_FLOAT_EQ(metric->Evaluate(preds, p_fmat), 0.50f);
info.labels_upper_bound_.HostVector()[2] = std::numeric_limits<bst_float>::infinity();
EXPECT_FLOAT_EQ(metric->Eval(preds, info), 0.50f);
EXPECT_FLOAT_EQ(metric->Evaluate(preds, p_fmat), 0.50f);
info.labels_upper_bound_.HostVector()[3] = std::numeric_limits<bst_float>::infinity();
EXPECT_FLOAT_EQ(metric->Eval(preds, info), 0.50f);
EXPECT_FLOAT_EQ(metric->Evaluate(preds, p_fmat), 0.50f);
info.labels_lower_bound_.HostVector()[0] = 70.0f;
EXPECT_FLOAT_EQ(metric->Eval(preds, info), 0.25f);
EXPECT_FLOAT_EQ(metric->Evaluate(preds, p_fmat), 0.25f);
CheckDeterministicMetricElementWise(StringView{"interval-regression-accuracy"}, GPUIDX);
}