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

Update GPUTreeshap (#6163)

Browse Source 
* Reduce shap test duration

* Test interoperability with shap package

* Add feature interactions

* Update GPUTreeShap
This commit is contained in:
Rory Mitchell
2020-09-28 09:43:47 +13:00
committed by GitHub
parent 434a3f35a3
commit dda9e1e487
14 changed files with 176 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
tests/cpp/predictor/test_cpu_predictor.cc
View File

@@ -53,24 +53,28 @@ TEST(CpuPredictor, Basic) {
}
// Test predict contribution
std::vector<float> out_contribution;
cpu_predictor->PredictContribution(dmat.get(), &out_contribution, model);
HostDeviceVector<float> out_contribution_hdv;
auto& out_contribution = out_contribution_hdv.HostVector();
cpu_predictor->PredictContribution(dmat.get(), &out_contribution_hdv, model);
ASSERT_EQ(out_contribution.size(), kRows * (kCols + 1));
for (size_t i = 0; i < out_contribution.size(); ++i) {
auto const& contri = out_contribution[i];
// shift 1 for bias, as test tree is a decision dump, only global bias is filled with LeafValue().
if ((i+1) % (kCols+1) == 0) {
// shift 1 for bias, as test tree is a decision dump, only global bias is
// filled with LeafValue().
if ((i + 1) % (kCols + 1) == 0) {
ASSERT_EQ(out_contribution.back(), 1.5f);
} else {
ASSERT_EQ(contri, 0);
}
}
// Test predict contribution (approximate method)
cpu_predictor->PredictContribution(dmat.get(), &out_contribution, model, 0, nullptr, true);
cpu_predictor->PredictContribution(dmat.get(), &out_contribution_hdv, model,
0, nullptr, true);
for (size_t i = 0; i < out_contribution.size(); ++i) {
auto const& contri = out_contribution[i];
// shift 1 for bias, as test tree is a decision dump, only global bias is filled with LeafValue().
if ((i+1) % (kCols+1) == 0) {
// shift 1 for bias, as test tree is a decision dump, only global bias is
// filled with LeafValue().
if ((i + 1) % (kCols + 1) == 0) {
ASSERT_EQ(out_contribution.back(), 1.5f);
} else {
ASSERT_EQ(contri, 0);
@@ -112,8 +116,9 @@ TEST(CpuPredictor, ExternalMemory) {
}
// Test predict contribution
std::vector<float> out_contribution;
cpu_predictor->PredictContribution(dmat.get(), &out_contribution, model);
HostDeviceVector<float> out_contribution_hdv;
auto& out_contribution = out_contribution_hdv.HostVector();
cpu_predictor->PredictContribution(dmat.get(), &out_contribution_hdv, model);
ASSERT_EQ(out_contribution.size(), dmat->Info().num_row_ * (dmat->Info().num_col_ + 1));
for (size_t i = 0; i < out_contribution.size(); ++i) {
auto const& contri = out_contribution[i];
@@ -126,8 +131,10 @@ TEST(CpuPredictor, ExternalMemory) {
}
// Test predict contribution (approximate method)
std::vector<float> out_contribution_approximate;
cpu_predictor->PredictContribution(dmat.get(), &out_contribution_approximate, model, 0, nullptr, true);
HostDeviceVector<float> out_contribution_approximate_hdv;
auto& out_contribution_approximate = out_contribution_approximate_hdv.HostVector();
cpu_predictor->PredictContribution(
dmat.get(), &out_contribution_approximate_hdv, model, 0, nullptr, true);
ASSERT_EQ(out_contribution_approximate.size(),
dmat->Info().num_row_ * (dmat->Info().num_col_ + 1));
for (size_t i = 0; i < out_contribution.size(); ++i) {

32
tests/cpp/predictor/test_gpu_predictor.cu
View File

@@ -176,12 +176,13 @@ TEST(GPUPredictor, ShapStump) {
model.CommitModel(std::move(trees), 0);
auto gpu_lparam = CreateEmptyGenericParam(0);
std::unique_ptr<Predictor> gpu_predictor =
std::unique_ptr<Predictor>(Predictor::Create("gpu_predictor", &gpu_lparam));
std::unique_ptr<Predictor> gpu_predictor = std::unique_ptr<Predictor>(
Predictor::Create("gpu_predictor", &gpu_lparam));
gpu_predictor->Configure({});
std::vector<float > phis;
auto dmat = RandomDataGenerator(3, 1, 0).GenerateDMatrix();
gpu_predictor->PredictContribution(dmat.get(), &phis, model);
HostDeviceVector<float> predictions;
auto dmat = RandomDataGenerator(3, 1, 0).GenerateDMatrix();
gpu_predictor->PredictContribution(dmat.get(), &predictions, model);
auto& phis = predictions.HostVector();
EXPECT_EQ(phis[0], 0.0);
EXPECT_EQ(phis[1], param.base_score);
EXPECT_EQ(phis[2], 0.0);
@@ -202,19 +203,20 @@ TEST(GPUPredictor, Shap) {
auto gpu_lparam = CreateEmptyGenericParam(0);
auto cpu_lparam = CreateEmptyGenericParam(-1);
std::unique_ptr<Predictor> gpu_predictor =
std::unique_ptr<Predictor>(Predictor::Create("gpu_predictor", &gpu_lparam));
std::unique_ptr<Predictor> cpu_predictor =
std::unique_ptr<Predictor>(Predictor::Create("cpu_predictor", &cpu_lparam));
std::unique_ptr<Predictor> gpu_predictor = std::unique_ptr<Predictor>(
Predictor::Create("gpu_predictor", &gpu_lparam));
std::unique_ptr<Predictor> cpu_predictor = std::unique_ptr<Predictor>(
Predictor::Create("cpu_predictor", &cpu_lparam));
gpu_predictor->Configure({});
cpu_predictor->Configure({});
std::vector<float > phis;
std::vector<float > cpu_phis;
HostDeviceVector<float> predictions;
HostDeviceVector<float> cpu_predictions;
auto dmat = RandomDataGenerator(3, 1, 0).GenerateDMatrix();
gpu_predictor->PredictContribution(dmat.get(), &phis, model);
cpu_predictor->PredictContribution(dmat.get(), &cpu_phis, model);
for(auto i = 0ull; i < phis.size(); i++)
{
gpu_predictor->PredictContribution(dmat.get(), &predictions, model);
cpu_predictor->PredictContribution(dmat.get(), &cpu_predictions, model);
auto& phis = predictions.HostVector();
auto& cpu_phis = cpu_predictions.HostVector();
for (auto i = 0ull; i < phis.size(); i++) {
EXPECT_NEAR(cpu_phis[i], phis[i], 1e-3);
}
}