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

GPU implementation of AFT survival objective and metric (#5714)

Browse Source 
* Add interval accuracy

* De-virtualize AFT functions

* Lint

* Refactor AFT metric using GPU-CPU reducer

* Fix R build

* Fix build on Windows

* Fix copyright header

* Clang-tidy

* Fix crashing demo

* Fix typos in comment; explain GPU ID

* Remove unnecessary #include

* Add C++ test for interval accuracy

* Fix a bug in accuracy metric: use log pred

* Refactor AFT objective using GPU-CPU Transform

* Lint

* Fix lint

* Use Ninja to speed up build

* Use time, not /usr/bin/time

* Add cpu_build worker class, with concurrency = 1

* Use concurrency = 1 only for CUDA build

* concurrency = 1 for clang-tidy

* Address reviewer's feedback

* Update link to AFT paper
This commit is contained in:
Philip Hyunsu Cho
2020-07-17 01:18:13 -07:00
committed by GitHub
parent 7c2686146e
commit 71b0528a2f
20 changed files with 1050 additions and 822 deletions
Download Patch File Download Diff File Expand all files Collapse all files

100
tests/cpp/common/test_probability_distribution.cc
View File

@@ -11,59 +11,56 @@
namespace xgboost {
namespace common {
TEST(ProbabilityDistribution, DistributionGeneric) {
// Assert d/dx CDF = PDF, d/dx PDF = GradPDF, d/dx GradPDF = HessPDF
// Do this for every distribution type
for (auto type : {ProbabilityDistributionType::kNormal, ProbabilityDistributionType::kLogistic,
ProbabilityDistributionType::kExtreme}) {
std::unique_ptr<ProbabilityDistribution> dist{ ProbabilityDistribution::Create(type) };
double integral_of_pdf = dist->CDF(-2.0);
double integral_of_grad_pdf = dist->PDF(-2.0);
double integral_of_hess_pdf = dist->GradPDF(-2.0);
// Perform numerical differentiation and integration
// Enumerate 4000 grid points in range [-2, 2]
for (int i = 0; i <= 4000; ++i) {
const double x = static_cast<double>(i) / 1000.0 - 2.0;
// Numerical differentiation (p. 246, Numerical Analysis 2nd ed. by Timothy Sauer)
EXPECT_NEAR((dist->CDF(x + 1e-5) - dist->CDF(x - 1e-5)) / 2e-5, dist->PDF(x), 6e-11);
EXPECT_NEAR((dist->PDF(x + 1e-5) - dist->PDF(x - 1e-5)) / 2e-5, dist->GradPDF(x), 6e-11);
EXPECT_NEAR((dist->GradPDF(x + 1e-5) - dist->GradPDF(x - 1e-5)) / 2e-5,
dist->HessPDF(x), 6e-11);
// Numerical integration using Trapezoid Rule (p. 257, Sauer)
integral_of_pdf += 5e-4 * (dist->PDF(x - 1e-3) + dist->PDF(x));
integral_of_grad_pdf += 5e-4 * (dist->GradPDF(x - 1e-3) + dist->GradPDF(x));
integral_of_hess_pdf += 5e-4 * (dist->HessPDF(x - 1e-3) + dist->HessPDF(x));
EXPECT_NEAR(integral_of_pdf, dist->CDF(x), 2e-4);
EXPECT_NEAR(integral_of_grad_pdf, dist->PDF(x), 2e-4);
EXPECT_NEAR(integral_of_hess_pdf, dist->GradPDF(x), 2e-4);
}
template <typename Distribution>
void RunDistributionGenericTest() {
double integral_of_pdf = Distribution::CDF(-2.0);
double integral_of_grad_pdf = Distribution::PDF(-2.0);
double integral_of_hess_pdf = Distribution::GradPDF(-2.0);
// Perform numerical differentiation and integration
// Enumerate 4000 grid points in range [-2, 2]
for (int i = 0; i <= 4000; ++i) {
const double x = static_cast<double>(i) / 1000.0 - 2.0;
// Numerical differentiation (p. 246, Numerical Analysis 2nd ed. by Timothy Sauer)
EXPECT_NEAR((Distribution::CDF(x + 1e-5) - Distribution::CDF(x - 1e-5)) / 2e-5,
Distribution::PDF(x), 6e-11);
EXPECT_NEAR((Distribution::PDF(x + 1e-5) - Distribution::PDF(x - 1e-5)) / 2e-5,
Distribution::GradPDF(x), 6e-11);
EXPECT_NEAR((Distribution::GradPDF(x + 1e-5) - Distribution::GradPDF(x - 1e-5)) / 2e-5,
Distribution::HessPDF(x), 6e-11);
// Numerical integration using Trapezoid Rule (p. 257, Sauer)
integral_of_pdf += 5e-4 * (Distribution::PDF(x - 1e-3) + Distribution::PDF(x));
integral_of_grad_pdf += 5e-4 * (Distribution::GradPDF(x - 1e-3) + Distribution::GradPDF(x));
integral_of_hess_pdf += 5e-4 * (Distribution::HessPDF(x - 1e-3) + Distribution::HessPDF(x));
EXPECT_NEAR(integral_of_pdf, Distribution::CDF(x), 2e-4);
EXPECT_NEAR(integral_of_grad_pdf, Distribution::PDF(x), 2e-4);
EXPECT_NEAR(integral_of_hess_pdf, Distribution::GradPDF(x), 2e-4);
}
}
TEST(ProbabilityDistribution, NormalDist) {
std::unique_ptr<ProbabilityDistribution> dist{
ProbabilityDistribution::Create(ProbabilityDistributionType::kNormal)
};
TEST(ProbabilityDistribution, DistributionGeneric) {
// Assert d/dx CDF = PDF, d/dx PDF = GradPDF, d/dx GradPDF = HessPDF
// Do this for every distribution type
RunDistributionGenericTest<NormalDistribution>();
RunDistributionGenericTest<LogisticDistribution>();
RunDistributionGenericTest<ExtremeDistribution>();
}
TEST(ProbabilityDistribution, NormalDist) {
// "Three-sigma rule" (https://en.wikipedia.org/wiki/68–95–99.7_rule)
// 68% of values are within 1 standard deviation away from the mean
// 95% of values are within 2 standard deviation away from the mean
// 99.7% of values are within 3 standard deviation away from the mean
EXPECT_NEAR(dist->CDF(0.5) - dist->CDF(-0.5), 0.3829, 0.00005);
EXPECT_NEAR(dist->CDF(1.0) - dist->CDF(-1.0), 0.6827, 0.00005);
EXPECT_NEAR(dist->CDF(1.5) - dist->CDF(-1.5), 0.8664, 0.00005);
EXPECT_NEAR(dist->CDF(2.0) - dist->CDF(-2.0), 0.9545, 0.00005);
EXPECT_NEAR(dist->CDF(2.5) - dist->CDF(-2.5), 0.9876, 0.00005);
EXPECT_NEAR(dist->CDF(3.0) - dist->CDF(-3.0), 0.9973, 0.00005);
EXPECT_NEAR(dist->CDF(3.5) - dist->CDF(-3.5), 0.9995, 0.00005);
EXPECT_NEAR(dist->CDF(4.0) - dist->CDF(-4.0), 0.9999, 0.00005);
EXPECT_NEAR(NormalDistribution::CDF(0.5) - NormalDistribution::CDF(-0.5), 0.3829, 0.00005);
EXPECT_NEAR(NormalDistribution::CDF(1.0) - NormalDistribution::CDF(-1.0), 0.6827, 0.00005);
EXPECT_NEAR(NormalDistribution::CDF(1.5) - NormalDistribution::CDF(-1.5), 0.8664, 0.00005);
EXPECT_NEAR(NormalDistribution::CDF(2.0) - NormalDistribution::CDF(-2.0), 0.9545, 0.00005);
EXPECT_NEAR(NormalDistribution::CDF(2.5) - NormalDistribution::CDF(-2.5), 0.9876, 0.00005);
EXPECT_NEAR(NormalDistribution::CDF(3.0) - NormalDistribution::CDF(-3.0), 0.9973, 0.00005);
EXPECT_NEAR(NormalDistribution::CDF(3.5) - NormalDistribution::CDF(-3.5), 0.9995, 0.00005);
EXPECT_NEAR(NormalDistribution::CDF(4.0) - NormalDistribution::CDF(-4.0), 0.9999, 0.00005);
}
TEST(ProbabilityDistribution, LogisticDist) {
std::unique_ptr<ProbabilityDistribution> dist{
ProbabilityDistribution::Create(ProbabilityDistributionType::kLogistic)
};
/**
* Enforce known properties of the logistic distribution.
* (https://en.wikipedia.org/wiki/Logistic_distribution)
@@ -74,17 +71,13 @@ TEST(ProbabilityDistribution, LogisticDist) {
const double x = static_cast<double>(i) / 1000.0 - 2.0;
// PDF = 1/4 * sech(x/2)**2
const double sech_x = 1.0 / std::cosh(x * 0.5); // hyperbolic secant at x/2
EXPECT_NEAR(0.25 * sech_x * sech_x, dist->PDF(x), 1e-15);
EXPECT_NEAR(0.25 * sech_x * sech_x, LogisticDistribution::PDF(x), 1e-15);
// CDF = 1/2 + 1/2 * tanh(x/2)
EXPECT_NEAR(0.5 + 0.5 * std::tanh(x * 0.5), dist->CDF(x), 1e-15);
EXPECT_NEAR(0.5 + 0.5 * std::tanh(x * 0.5), LogisticDistribution::CDF(x), 1e-15);
}
}
TEST(ProbabilityDistribution, ExtremeDist) {
std::unique_ptr<ProbabilityDistribution> dist{
ProbabilityDistribution::Create(ProbabilityDistributionType::kExtreme)
};
/**
* Enforce known properties of the extreme distribution (also known as Gumbel distribution).
* The mean is the negative of the Euler-Mascheroni constant.
@@ -99,9 +92,10 @@ TEST(ProbabilityDistribution, ExtremeDist) {
for (int i = 0; i <= 25000; ++i) {
const double x = static_cast<double>(i) / 1000.0 - 20.0;
// Numerical integration using Trapezoid Rule (p. 257, Sauer)
mean += 5e-4 * ((x - 1e-3) * dist->PDF(x - 1e-3) + x * dist->PDF(x));
mean +=
5e-4 * ((x - 1e-3) * ExtremeDistribution::PDF(x - 1e-3) + x * ExtremeDistribution::PDF(x));
}
EXPECT_NEAR(mean, -probability_constant::kEulerMascheroni, 1e-7);
EXPECT_NEAR(mean, -kEulerMascheroni, 1e-7);
// Enumerate 25000 grid points in range [-20, 5].
// Compute the variance of the distribution using numerical integration.
@@ -111,10 +105,10 @@ TEST(ProbabilityDistribution, ExtremeDist) {
for (int i = 0; i <= 25000; ++i) {
const double x = static_cast<double>(i) / 1000.0 - 20.0;
// Numerical integration using Trapezoid Rule (p. 257, Sauer)
variance += 5e-4 * ((x - 1e-3 - mean) * (x - 1e-3 - mean) * dist->PDF(x - 1e-3)
+ (x - mean) * (x - mean) * dist->PDF(x));
variance += 5e-4 * ((x - 1e-3 - mean) * (x - 1e-3 - mean) * ExtremeDistribution::PDF(x - 1e-3)
+ (x - mean) * (x - mean) * ExtremeDistribution::PDF(x));
}
EXPECT_NEAR(variance, probability_constant::kPI * probability_constant::kPI / 6.0, 1e-6);
EXPECT_NEAR(variance, kPI * kPI / 6.0, 1e-6);
}
} // namespace common

31
tests/cpp/common/test_survival_util.cc
View File

@@ -8,36 +8,37 @@
namespace xgboost {
namespace common {
inline static void RobustTestSuite(ProbabilityDistributionType dist_type,
double y_lower, double y_upper, double sigma) {
AFTLoss loss(dist_type);
template <typename Distribution>
inline static void RobustTestSuite(double y_lower, double y_upper, double sigma) {
for (int i = 50; i >= -50; --i) {
const double y_pred = std::pow(10.0, static_cast<double>(i));
const double z = (std::log(y_lower) - std::log(y_pred)) / sigma;
const double gradient = loss.Gradient(y_lower, y_upper, std::log(y_pred), sigma);
const double hessian = loss.Hessian(y_lower, y_upper, std::log(y_pred), sigma);
const double gradient
= AFTLoss<Distribution>::Gradient(y_lower, y_upper, std::log(y_pred), sigma);
const double hessian
= AFTLoss<Distribution>::Hessian(y_lower, y_upper, std::log(y_pred), sigma);
ASSERT_FALSE(std::isnan(gradient)) << "z = " << z << ", y \\in ["
<< y_lower << ", " << y_upper << "], y_pred = " << y_pred
<< ", dist = " << static_cast<int>(dist_type);
<< ", dist = " << static_cast<int>(Distribution::Type());
ASSERT_FALSE(std::isinf(gradient)) << "z = " << z << ", y \\in ["
<< y_lower << ", " << y_upper << "], y_pred = " << y_pred
<< ", dist = " << static_cast<int>(dist_type);
<< ", dist = " << static_cast<int>(Distribution::Type());
ASSERT_FALSE(std::isnan(hessian)) << "z = " << z << ", y \\in ["
<< y_lower << ", " << y_upper << "], y_pred = " << y_pred
<< ", dist = " << static_cast<int>(dist_type);
<< ", dist = " << static_cast<int>(Distribution::Type());
ASSERT_FALSE(std::isinf(hessian)) << "z = " << z << ", y \\in ["
<< y_lower << ", " << y_upper << "], y_pred = " << y_pred
<< ", dist = " << static_cast<int>(dist_type);
<< ", dist = " << static_cast<int>(Distribution::Type());
}
}
TEST(AFTLoss, RobustGradientPair) { // Ensure that INF and NAN don't show up in gradient pair
RobustTestSuite(ProbabilityDistributionType::kNormal, 16.0, 200.0, 2.0);
RobustTestSuite(ProbabilityDistributionType::kLogistic, 16.0, 200.0, 2.0);
RobustTestSuite(ProbabilityDistributionType::kExtreme, 16.0, 200.0, 2.0);
RobustTestSuite(ProbabilityDistributionType::kNormal, 100.0, 100.0, 2.0);
RobustTestSuite(ProbabilityDistributionType::kLogistic, 100.0, 100.0, 2.0);
RobustTestSuite(ProbabilityDistributionType::kExtreme, 100.0, 100.0, 2.0);
RobustTestSuite<NormalDistribution>(16.0, 200.0, 2.0);
RobustTestSuite<LogisticDistribution>(16.0, 200.0, 2.0);
RobustTestSuite<ExtremeDistribution>(16.0, 200.0, 2.0);
RobustTestSuite<NormalDistribution>(100.0, 100.0, 2.0);
RobustTestSuite<LogisticDistribution>(100.0, 100.0, 2.0);
RobustTestSuite<ExtremeDistribution>(100.0, 100.0, 2.0);
}
} // namespace common

95
tests/cpp/metric/test_survival_metric.cc
View File

@@ -13,78 +13,39 @@
#include "../helpers.h"
#include "../../../src/common/survival_util.h"
// CUDA conditional compile trick.
#include "test_survival_metric.cu"
namespace xgboost {
namespace common {
/** Tests for Survival metrics that should run only on CPU **/
/**
* Reference values obtained from
* https://github.com/avinashbarnwal/GSOC-2019/blob/master/AFT/R/combined_assignment.R
**/
TEST(Metric, AFTNegLogLik) {
auto lparam = CreateEmptyGenericParam(-1); // currently AFT metric is CPU only
/**
* 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;
info.num_row_ = 4;
info.labels_lower_bound_.HostVector()
= { 100.0f, -std::numeric_limits<bst_float>::infinity(), 60.0f, 16.0f };
info.labels_upper_bound_.HostVector()
= { 100.0f, 20.0f, std::numeric_limits<bst_float>::infinity(), 200.0f };
info.weights_.HostVector() = std::vector<bst_float>();
HostDeviceVector<bst_float> preds(4, std::log(64));
struct TestCase {
std::string dist_type;
bst_float reference_value;
};
for (const auto& test_case : std::vector<TestCase>{ {"normal", 2.1508f}, {"logistic", 2.1804f},
{"extreme", 2.0706f} }) {
std::unique_ptr<Metric> metric(Metric::Create("aft-nloglik", &lparam));
metric->Configure({ {"aft_loss_distribution", test_case.dist_type},
{"aft_loss_distribution_scale", "1.0"} });
EXPECT_NEAR(metric->Eval(preds, info, false), test_case.reference_value, 1e-4);
}
}
// Test configuration of AFT metric
TEST(AFTNegLogLikMetric, Configuration) {
auto lparam = CreateEmptyGenericParam(-1); // currently AFT metric is CPU only
std::unique_ptr<Metric> metric(Metric::Create("aft-nloglik", &lparam));
metric->Configure({{"aft_loss_distribution", "normal"}, {"aft_loss_distribution_scale", "10"}});
// Configuration round-trip test
Json j_obj{ Object() };
metric->SaveConfig(&j_obj);
auto aft_param_json = j_obj["aft_loss_param"];
EXPECT_EQ(get<String>(aft_param_json["aft_loss_distribution"]), "normal");
EXPECT_EQ(get<String>(aft_param_json["aft_loss_distribution_scale"]), "10");
}
/**
* AFTLoss.* tests verify metric values over individual data points.
**/
// Generate prediction value ranging from 2**1 to 2**15, using grid points in log scale
// Then check prediction against the reference values
template <typename Distribution>
static inline void CheckLossOverGridPoints(
double true_label_lower_bound,
double true_label_upper_bound,
ProbabilityDistributionType dist_type,
const std::vector<double>& reference_values) {
const int num_point = 20;
const double log_y_low = 1.0;
const double log_y_high = 15.0;
std::unique_ptr<AFTLoss> loss(new AFTLoss(dist_type));
CHECK_EQ(num_point, reference_values.size());
for (int i = 0; i < num_point; ++i) {
const double y_pred
= std::pow(2.0, i * (log_y_high - log_y_low) / (num_point - 1) + log_y_low);
const double loss_val
= loss->Loss(true_label_lower_bound, true_label_upper_bound, std::log(y_pred), 1.0);
const double loss_val = AFTLoss<Distribution>::Loss(
true_label_lower_bound, true_label_upper_bound, std::log(y_pred), 1.0);
EXPECT_NEAR(loss_val, reference_values[i], 1e-4);
}
}
@@ -94,35 +55,29 @@ TEST(AFTLoss, Uncensored) {
const double true_label_lower_bound = 100.0;
const double true_label_upper_bound = true_label_lower_bound;
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kNormal,
CheckLossOverGridPoints<NormalDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 13.1761, 11.3085, 9.7017, 8.3558, 7.2708, 6.4466, 5.8833, 5.5808, 5.5392, 5.7585, 6.2386,
6.9795, 7.9813, 9.2440, 10.7675, 12.5519, 14.5971, 16.9032, 19.4702, 22.2980 });
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kLogistic,
CheckLossOverGridPoints<LogisticDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 8.5568, 8.0720, 7.6038, 7.1620, 6.7612, 6.4211, 6.1659, 6.0197, 5.9990, 6.1064, 6.3293,
6.6450, 7.0289, 7.4594, 7.9205, 8.4008, 8.8930, 9.3926, 9.8966, 10.4033 });
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kExtreme,
CheckLossOverGridPoints<ExtremeDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 27.6310, 27.6310, 19.7177, 13.0281, 9.2183, 7.1365, 6.0916, 5.6688, 5.6195, 5.7941, 6.1031,
6.4929, 6.9310, 7.3981, 7.8827, 8.3778, 8.8791, 9.3842, 9.8916, 10.40033 });
}
TEST(AFTLoss, LeftCensored) {
// Given label (-inf, 20], compute the AFT loss for various prediction values
const double true_label_lower_bound = -std::numeric_limits<double>::infinity();
const double true_label_lower_bound = 0.0;
const double true_label_upper_bound = 20.0;
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kNormal,
CheckLossOverGridPoints<NormalDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 0.0107, 0.0373, 0.1054, 0.2492, 0.5068, 0.9141, 1.5003, 2.2869, 3.2897, 4.5196, 5.9846,
7.6902, 9.6405, 11.8385, 14.2867, 16.9867, 19.9399, 23.1475, 26.6103, 27.6310 });
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kLogistic,
CheckLossOverGridPoints<LogisticDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 0.0953, 0.1541, 0.2451, 0.3804, 0.5717, 0.8266, 1.1449, 1.5195, 1.9387, 2.3902, 2.8636,
3.3512, 3.8479, 4.3500, 4.8556, 5.3632, 5.8721, 6.3817, 6.8918, 7.4021 });
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kExtreme,
CheckLossOverGridPoints<ExtremeDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 0.0000, 0.0025, 0.0277, 0.1225, 0.3195, 0.6150, 0.9862, 1.4094, 1.8662, 2.3441, 2.8349,
3.3337, 3.8372, 4.3436, 4.8517, 5.3609, 5.8707, 6.3808, 6.8912, 7.4018 });
}
@@ -132,16 +87,13 @@ TEST(AFTLoss, RightCensored) {
const double true_label_lower_bound = 60.0;
const double true_label_upper_bound = std::numeric_limits<double>::infinity();
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kNormal,
CheckLossOverGridPoints<NormalDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 8.0000, 6.2537, 4.7487, 3.4798, 2.4396, 1.6177, 0.9993, 0.5638, 0.2834, 0.1232, 0.0450,
0.0134, 0.0032, 0.0006, 0.0001, 0.0000, 0.0000, 0.0000, 0.0000, 0.0000 });
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kLogistic,
CheckLossOverGridPoints<LogisticDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 3.4340, 2.9445, 2.4683, 2.0125, 1.5871, 1.2041, 0.8756, 0.6099, 0.4083, 0.2643, 0.1668,
0.1034, 0.0633, 0.0385, 0.0233, 0.0140, 0.0084, 0.0051, 0.0030, 0.0018 });
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kExtreme,
CheckLossOverGridPoints<ExtremeDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 27.6310, 18.0015, 10.8018, 6.4817, 3.8893, 2.3338, 1.4004, 0.8403, 0.5042, 0.3026, 0.1816,
0.1089, 0.0654, 0.0392, 0.0235, 0.0141, 0.0085, 0.0051, 0.0031, 0.0018 });
}
@@ -150,17 +102,14 @@ TEST(AFTLoss, IntervalCensored) {
// Given label [16, 200], compute the AFT loss for various prediction values
const double true_label_lower_bound = 16.0;
const double true_label_upper_bound = 200.0;
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kNormal,
CheckLossOverGridPoints<NormalDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 3.9746, 2.8415, 1.9319, 1.2342, 0.7335, 0.4121, 0.2536, 0.2470, 0.3919, 0.6982, 1.1825,
1.8622, 2.7526, 3.8656, 5.2102, 6.7928, 8.6183, 10.6901, 13.0108, 15.5826 });
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kLogistic,
CheckLossOverGridPoints<LogisticDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 2.2906, 1.8578, 1.4667, 1.1324, 0.8692, 0.6882, 0.5948, 0.5909, 0.6764, 0.8499, 1.1061,
1.4348, 1.8215, 2.2511, 2.7104, 3.1891, 3.6802, 4.1790, 4.6825, 5.1888 });
CheckLossOverGridPoints(true_label_lower_bound, true_label_upper_bound,
ProbabilityDistributionType::kExtreme,
CheckLossOverGridPoints<ExtremeDistribution>(true_label_lower_bound, true_label_upper_bound,
{ 8.0000, 4.8004, 2.8805, 1.7284, 1.0372, 0.6231, 0.3872, 0.3031, 0.3740, 0.5839, 0.8995,
1.2878, 1.7231, 2.1878, 2.6707, 3.1647, 3.6653, 4.1699, 4.6770, 5.1856 });
}

81
tests/cpp/metric/test_survival_metric.cu Normal file
View File

@@ -0,0 +1,81 @@
/*!
* Copyright (c) by Contributors 2020
*/
#include <gtest/gtest.h>
#include <cmath>
#include "xgboost/metric.h"
#include "../helpers.h"
#include "../../../src/common/survival_util.h"
/** Tests for Survival metrics that should run both on CPU and GPU **/
namespace xgboost {
namespace common {
TEST(Metric, DeclareUnifiedTest(AFTNegLogLik)) {
auto lparam = xgboost::CreateEmptyGenericParam(GPUIDX);
/**
* 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;
info.num_row_ = 4;
info.labels_lower_bound_.HostVector()
= { 100.0f, 0.0f, 60.0f, 16.0f };
info.labels_upper_bound_.HostVector()
= { 100.0f, 20.0f, std::numeric_limits<bst_float>::infinity(), 200.0f };
info.weights_.HostVector() = std::vector<bst_float>();
HostDeviceVector<bst_float> preds(4, std::log(64));
struct TestCase {
std::string dist_type;
bst_float reference_value;
};
for (const auto& test_case : std::vector<TestCase>{ {"normal", 2.1508f}, {"logistic", 2.1804f},
{"extreme", 2.0706f} }) {
std::unique_ptr<Metric> metric(Metric::Create("aft-nloglik", &lparam));
metric->Configure({ {"aft_loss_distribution", test_case.dist_type},
{"aft_loss_distribution_scale", "1.0"} });
EXPECT_NEAR(metric->Eval(preds, info, false), test_case.reference_value, 1e-4);
}
}
TEST(Metric, DeclareUnifiedTest(IntervalRegressionAccuracy)) {
auto lparam = xgboost::CreateEmptyGenericParam(GPUIDX);
MetaInfo 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 };
info.weights_.HostVector() = std::vector<bst_float>();
HostDeviceVector<bst_float> preds(4, std::log(60.0f));
std::unique_ptr<Metric> metric(Metric::Create("interval-regression-accuracy", &lparam));
EXPECT_FLOAT_EQ(metric->Eval(preds, info, false), 0.75f);
info.labels_lower_bound_.HostVector()[2] = 70.0f;
EXPECT_FLOAT_EQ(metric->Eval(preds, info, false), 0.50f);
info.labels_upper_bound_.HostVector()[2] = std::numeric_limits<bst_float>::infinity();
EXPECT_FLOAT_EQ(metric->Eval(preds, info, false), 0.50f);
info.labels_upper_bound_.HostVector()[3] = std::numeric_limits<bst_float>::infinity();
EXPECT_FLOAT_EQ(metric->Eval(preds, info, false), 0.50f);
info.labels_lower_bound_.HostVector()[0] = 70.0f;
EXPECT_FLOAT_EQ(metric->Eval(preds, info, false), 0.25f);
}
// Test configuration of AFT metric
TEST(AFTNegLogLikMetric, DeclareUnifiedTest(Configuration)) {
auto lparam = xgboost::CreateEmptyGenericParam(GPUIDX);
std::unique_ptr<Metric> metric(Metric::Create("aft-nloglik", &lparam));
metric->Configure({{"aft_loss_distribution", "normal"}, {"aft_loss_distribution_scale", "10"}});
// Configuration round-trip test
Json j_obj{ Object() };
metric->SaveConfig(&j_obj);
auto aft_param_json = j_obj["aft_loss_param"];
EXPECT_EQ(get<String>(aft_param_json["aft_loss_distribution"]), "normal");
EXPECT_EQ(get<String>(aft_param_json["aft_loss_distribution_scale"]), "10");
}
} // namespace common
} // namespace xgboost

26
tests/cpp/objective/test_aft_obj.cc
View File

@@ -15,8 +15,8 @@
namespace xgboost {
namespace common {
TEST(Objective, AFTObjConfiguration) {
auto lparam = CreateEmptyGenericParam(-1); // currently AFT objective is CPU only
TEST(Objective, DeclareUnifiedTest(AFTObjConfiguration)) {
auto lparam = CreateEmptyGenericParam(GPUIDX);
std::unique_ptr<ObjFunction> objective(ObjFunction::Create("survival:aft", &lparam));
objective->Configure({ {"aft_loss_distribution", "logistic"},
{"aft_loss_distribution_scale", "5"} });
@@ -76,8 +76,8 @@ static inline void CheckGPairOverGridPoints(
}
}
TEST(Objective, AFTObjGPairUncensoredLabels) {
auto lparam = CreateEmptyGenericParam(-1); // currently AFT objective is CPU only
TEST(Objective, DeclareUnifiedTest(AFTObjGPairUncensoredLabels)) {
auto lparam = CreateEmptyGenericParam(GPUIDX);
std::unique_ptr<ObjFunction> obj(ObjFunction::Create("survival:aft", &lparam));
CheckGPairOverGridPoints(obj.get(), 100.0f, 100.0f, "normal",
@@ -100,29 +100,29 @@ TEST(Objective, AFTObjGPairUncensoredLabels) {
0.3026f, 0.1816f, 0.1090f, 0.0654f, 0.0392f, 0.0235f, 0.0141f, 0.0085f, 0.0051f, 0.0031f });
}
TEST(Objective, AFTObjGPairLeftCensoredLabels) {
auto lparam = CreateEmptyGenericParam(-1); // currently AFT objective is CPU only
TEST(Objective, DeclareUnifiedTest(AFTObjGPairLeftCensoredLabels)) {
auto lparam = CreateEmptyGenericParam(GPUIDX);
std::unique_ptr<ObjFunction> obj(ObjFunction::Create("survival:aft", &lparam));
CheckGPairOverGridPoints(obj.get(), -std::numeric_limits<float>::infinity(), 20.0f, "normal",
CheckGPairOverGridPoints(obj.get(), 0.0f, 20.0f, "normal",
{ 0.0285f, 0.0832f, 0.1951f, 0.3804f, 0.6403f, 0.9643f, 1.3379f, 1.7475f, 2.1828f, 2.6361f,
3.1023f, 3.5779f, 4.0603f, 4.5479f, 5.0394f, 5.5340f, 6.0309f, 6.5298f, 7.0303f, 7.5326f },
{ 0.0663f, 0.1559f, 0.2881f, 0.4378f, 0.5762f, 0.6878f, 0.7707f, 0.8300f, 0.8719f, 0.9016f,
0.9229f, 0.9385f, 0.9501f, 0.9588f, 0.9656f, 0.9709f, 0.9751f, 0.9785f, 0.9813f, 0.9877f });
CheckGPairOverGridPoints(obj.get(), -std::numeric_limits<float>::infinity(), 20.0f, "logistic",
CheckGPairOverGridPoints(obj.get(), 0.0f, 20.0f, "logistic",
{ 0.0909f, 0.1428f, 0.2174f, 0.3164f, 0.4355f, 0.5625f, 0.6818f, 0.7812f, 0.8561f, 0.9084f,
0.9429f, 0.9650f, 0.9787f, 0.9871f, 0.9922f, 0.9953f, 0.9972f, 0.9983f, 0.9990f, 0.9994f },
{ 0.0826f, 0.1224f, 0.1701f, 0.2163f, 0.2458f, 0.2461f, 0.2170f, 0.1709f, 0.1232f, 0.0832f,
0.0538f, 0.0338f, 0.0209f, 0.0127f, 0.0077f, 0.0047f, 0.0028f, 0.0017f, 0.0010f, 0.0006f });
CheckGPairOverGridPoints(obj.get(), -std::numeric_limits<float>::infinity(), 20.0f, "extreme",
CheckGPairOverGridPoints(obj.get(), 0.0f, 20.0f, "extreme",
{ 0.0005f, 0.0149f, 0.1011f, 0.2815f, 0.4881f, 0.6610f, 0.7847f, 0.8665f, 0.9183f, 0.9504f,
0.9700f, 0.9820f, 0.9891f, 0.9935f, 0.9961f, 0.9976f, 0.9986f, 0.9992f, 0.9995f, 0.9997f },
{ 0.0041f, 0.0747f, 0.2731f, 0.4059f, 0.3829f, 0.2901f, 0.1973f, 0.1270f, 0.0793f, 0.0487f,
0.0296f, 0.0179f, 0.0108f, 0.0065f, 0.0039f, 0.0024f, 0.0014f, 0.0008f, 0.0005f, 0.0003f });
}
TEST(Objective, AFTObjGPairRightCensoredLabels) {
auto lparam = CreateEmptyGenericParam(-1); // currently AFT objective is CPU only
TEST(Objective, DeclareUnifiedTest(AFTObjGPairRightCensoredLabels)) {
auto lparam = CreateEmptyGenericParam(GPUIDX);
std::unique_ptr<ObjFunction> obj(ObjFunction::Create("survival:aft", &lparam));
CheckGPairOverGridPoints(obj.get(), 60.0f, std::numeric_limits<float>::infinity(), "normal",
@@ -145,8 +145,8 @@ TEST(Objective, AFTObjGPairRightCensoredLabels) {
0.1816f, 0.1089f, 0.0654f, 0.0392f, 0.0235f, 0.0141f, 0.0085f, 0.0051f, 0.0031f, 0.0018f });
}
TEST(Objective, AFTObjGPairIntervalCensoredLabels) {
auto lparam = CreateEmptyGenericParam(-1); // currently AFT objective is CPU only
TEST(Objective, DeclareUnifiedTest(AFTObjGPairIntervalCensoredLabels)) {
auto lparam = CreateEmptyGenericParam(GPUIDX);
std::unique_ptr<ObjFunction> obj(ObjFunction::Create("survival:aft", &lparam));
CheckGPairOverGridPoints(obj.get(), 16.0f, 200.0f, "normal",

6
tests/cpp/objective/test_aft_obj.cu Normal file
View File

@@ -0,0 +1,6 @@
/*!
* Copyright 2020 XGBoost contributors
*/
// Dummy file to keep the CUDA tests.
#include "test_aft_obj.cc"