20211252547

Committed 14 Dec 2025 05:03PM UTC coverage: 67.712% (+30.3%) from 37.371%

Build # 20211252547

Build Type

push

github

Committed by

web-flow

Commit Message

Merge pull request #303 from libKriging/multistart-optim_jemalloc

Multistart optim

Coverage Stats

3946 of 4387 new or added lines in 30 files covered. (89.95%)

20 existing lines in 5 files now uncovered.

8403 of 12410 relevant lines covered (67.71%)

53457.78 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

97.92

/tests/NuggetKrigingUpdateTest.cpp

// clang-format off
// Must be first
#define CATCH_CONFIG_MAIN
#include "libKriging/utils/lk_armadillo.hpp"

#include <catch2/catch.hpp>
#include "libKriging/NuggetKriging.hpp"
// clang-format on

TEST_CASE("NuggetKrigingUpdateTest - Updated model equals combined fit", "[update][nuggetkriging]") {
  arma::arma_rng::set_seed(123);

  // Generate initial training data
  const arma::uword n_old = 10;
  const arma::uword n_new = 1;
  const arma::uword d = 2;
  
  arma::mat X_old(n_old, d, arma::fill::randu);
  arma::colvec y_old(n_old);
  
  // Use a simple test function with noise
  auto test_function = [](const arma::rowvec& x) {
    return x(0) * x(0) + x(1) * x(1);
  };
  
  for (arma::uword i = 0; i < n_old; ++i) {
    y_old(i) = test_function(X_old.row(i)) + 0.01 * arma::randn();
  }
  
  // Generate new data points
  arma::mat X_new(n_new, d, arma::fill::randu);
  arma::colvec y_new(n_new);
  
  for (arma::uword i = 0; i < n_new; ++i) {
    y_new(i) = test_function(X_new.row(i)) + 0.01 * arma::randn();
  }
  
  // Combine old and new data
  arma::mat X_combined = arma::join_cols(X_old, X_new);
  arma::colvec y_combined = arma::join_cols(y_old, y_new);

  SECTION("Update with refit equals combined fit") {
    // Fit model on old data
    NuggetKriging nk_updated("gauss");
    NuggetKriging::Parameters params{std::nullopt, true, std::nullopt, true, std::nullopt, true, std::nullopt, true};
    nk_updated.fit(y_old, X_old, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);
    
    // Update with new data
    nk_updated.update(y_new, X_new, true);
    
    // Fit model on combined data
    NuggetKriging nk_combined("gauss");
    nk_combined.fit(y_combined, X_combined, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);
    
    // Compare parameters
    INFO("Updated theta: " << nk_updated.theta().t());
    INFO("Combined theta: " << nk_combined.theta().t());
    INFO("Updated sigma2: " << nk_updated.sigma2());
    INFO("Combined sigma2: " << nk_combined.sigma2());
    INFO("Updated nugget: " << nk_updated.nugget());
    INFO("Combined nugget: " << nk_combined.nugget());
    INFO("Updated beta: " << nk_updated.beta().t());
    INFO("Combined beta: " << nk_combined.beta().t());
    
    // Check theta (should be very close) - using relative precision 0.05 (5%)
    // Use relative comparison, which handles zero values by falling back to absolute comparison
    CHECK(arma::approx_equal(nk_updated.theta(), nk_combined.theta(), "reldiff", 0.05));
    
    // Check sigma2 - using relative precision 0.05 (5%)
    if (nk_combined.sigma2() == 0.0) {
      CHECK(nk_updated.sigma2() == 0.0);  // If combined is zero, updated should also be zero
    } else {
      CHECK(std::abs(nk_updated.sigma2() - nk_combined.sigma2()) / std::abs(nk_combined.sigma2()) < 0.05);
    }
    
    // Check nugget - using relative precision 0.05 (5%)
    if (nk_combined.nugget() == 0.0) {
      CHECK(nk_updated.nugget() == 0.0);  // If combined is zero, updated should also be zero
    } else {
      CHECK(std::abs(nk_updated.nugget() - nk_combined.nugget()) / std::abs(nk_combined.nugget()) < 0.05);
    }
    
    // Check beta - using relative precision 0.05 (5%)
    // Use relative comparison, which handles zero values by falling back to absolute comparison
    CHECK(arma::approx_equal(nk_updated.beta(), nk_combined.beta(), "reldiff", 0.05));
    
    // Test predictions on new points (main validation)
    arma::mat X_test(5, d, arma::fill::randu);
    
    auto pred_updated = nk_updated.predict(X_test, true, false, false);
    auto pred_combined = nk_combined.predict(X_test, true, false, false);
    
    INFO("Max prediction difference: " << arma::max(arma::abs(std::get<0>(pred_updated) - std::get<0>(pred_combined))));
    
    CHECK(arma::approx_equal(std::get<0>(pred_updated), std::get<0>(pred_combined), "reldiff", 0.05));
  }

  SECTION("Multiple updates equal combined fit") {
    // Fit model on old data
    NuggetKriging nk_updated("gauss");
    NuggetKriging::Parameters params{std::nullopt, true, std::nullopt, true, std::nullopt, true, std::nullopt, true};
    nk_updated.fit(y_old, X_old, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);
    
    // Update one point at a time
    for (arma::uword i = 0; i < n_new; ++i) {
      arma::vec y_single = y_new.row(i);
      arma::mat X_single = X_new.row(i);
      nk_updated.update(y_single, X_single, true);
    }
    
    // Fit model on combined data
    NuggetKriging nk_combined("gauss");
    nk_combined.fit(y_combined, X_combined, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);
    
    // Test predictions
    arma::mat X_test(5, d, arma::fill::randu);
    
    auto pred_updated = nk_updated.predict(X_test, true, false, false);
    auto pred_combined = nk_combined.predict(X_test, true, false, false);
    
    INFO("Max prediction difference: " << arma::max(arma::abs(std::get<0>(pred_updated) - std::get<0>(pred_combined))));
    
    CHECK(arma::approx_equal(std::get<0>(pred_updated), std::get<0>(pred_combined), "reldiff", 0.05));
  }

  SECTION("Different kernels") {
    std::vector<std::string> kernels = {"gauss", "exp", "matern3_2", "matern5_2"};
    
    for (const auto& kernel : kernels) {
      INFO("Testing kernel: " << kernel);
      
      // Fit and update
      NuggetKriging nk_updated(kernel);
      NuggetKriging::Parameters params{std::nullopt, true, std::nullopt, true, std::nullopt, true, std::nullopt, true};
      nk_updated.fit(y_old, X_old, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);
      nk_updated.update(y_new, X_new, true);
      
      // Combined fit
      NuggetKriging nk_combined(kernel);
      nk_combined.fit(y_combined, X_combined, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);
      
      // Test predictions
      arma::mat X_test(3, d, arma::fill::randu);
      auto pred_updated = nk_updated.predict(X_test, true, false, false);
      auto pred_combined = nk_combined.predict(X_test, true, false, false);
      
      CHECK(arma::approx_equal(std::get<0>(pred_updated), std::get<0>(pred_combined), "reldiff", 0.05));
    }
  }

  SECTION("Different trend models") {
    std::vector<Trend::RegressionModel> trends = {
      Trend::RegressionModel::Constant,
      Trend::RegressionModel::Linear,
      Trend::RegressionModel::Quadratic
    };
    
    for (const auto& trend : trends) {
      INFO("Testing trend model");
      
      // Fit and update
      NuggetKriging nk_updated("gauss");
      NuggetKriging::Parameters params{std::nullopt, true, std::nullopt, true, std::nullopt, true, std::nullopt, true};
      nk_updated.fit(y_old, X_old, trend, false, "BFGS", "LL", params);
      nk_updated.update(y_new, X_new, true);
      
      // Combined fit
      NuggetKriging nk_combined("gauss");
      nk_combined.fit(y_combined, X_combined, trend, false, "BFGS", "LL", params);
      
      // Test predictions
      arma::mat X_test(3, d, arma::fill::randu);
      auto pred_updated = nk_updated.predict(X_test, true, false, false);
      auto pred_combined = nk_combined.predict(X_test, true, false, false);
      
      CHECK(arma::approx_equal(std::get<0>(pred_updated), std::get<0>(pred_combined), "reldiff", 0.05));
    }
  }
}

1	// clang-format off
2	// Must be first
3	#define CATCH_CONFIG_MAIN
4	#include "libKriging/utils/lk_armadillo.hpp"
5
6	#include <catch2/catch.hpp>
7	#include "libKriging/NuggetKriging.hpp"
8	// clang-format on
9
10	TEST_CASE("NuggetKrigingUpdateTest - Updated model equals combined fit", "[update][nuggetkriging]") {	4✔
11	arma::arma_rng::set_seed(123);	4✔
12
13	// Generate initial training data
14	const arma::uword n_old = 10;	4✔
15	const arma::uword n_new = 1;	4✔
16	const arma::uword d = 2;	4✔
17
18	arma::mat X_old(n_old, d, arma::fill::randu);	4✔
19	arma::colvec y_old(n_old);	4✔
20
21	// Use a simple test function with noise
22	auto test_function = [](const arma::rowvec& x) {	44✔
23	return x(0) * x(0) + x(1) * x(1);	220✔
24	};
25
26	for (arma::uword i = 0; i < n_old; ++i) {	44✔
27	y_old(i) = test_function(X_old.row(i)) + 0.01 * arma::randn();	80✔
28	}
29
30	// Generate new data points
31	arma::mat X_new(n_new, d, arma::fill::randu);	4✔
32	arma::colvec y_new(n_new);	4✔
33
34	for (arma::uword i = 0; i < n_new; ++i) {	8✔
35	y_new(i) = test_function(X_new.row(i)) + 0.01 * arma::randn();	8✔
36	}
37
38	// Combine old and new data
39	arma::mat X_combined = arma::join_cols(X_old, X_new);	4✔
40	arma::colvec y_combined = arma::join_cols(y_old, y_new);	4✔
41
42	SECTION("Update with refit equals combined fit") {	4✔
43	// Fit model on old data
44	NuggetKriging nk_updated("gauss");	2✔
45	NuggetKriging::Parameters params{std::nullopt, true, std::nullopt, true, std::nullopt, true, std::nullopt, true};	1✔
46	nk_updated.fit(y_old, X_old, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);	1✔
47
48	// Update with new data
49	nk_updated.update(y_new, X_new, true);	1✔
50
51	// Fit model on combined data
52	NuggetKriging nk_combined("gauss");	2✔
53	nk_combined.fit(y_combined, X_combined, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);	1✔
54
55	// Compare parameters
56	INFO("Updated theta: " << nk_updated.theta().t());	3✔
57	INFO("Combined theta: " << nk_combined.theta().t());	3✔
58	INFO("Updated sigma2: " << nk_updated.sigma2());	1✔
59	INFO("Combined sigma2: " << nk_combined.sigma2());	1✔
60	INFO("Updated nugget: " << nk_updated.nugget());	1✔
61	INFO("Combined nugget: " << nk_combined.nugget());	1✔
62	INFO("Updated beta: " << nk_updated.beta().t());	3✔
63	INFO("Combined beta: " << nk_combined.beta().t());	3✔
64
65	// Check theta (should be very close) - using relative precision 0.05 (5%)
66	// Use relative comparison, which handles zero values by falling back to absolute comparison
67	CHECK(arma::approx_equal(nk_updated.theta(), nk_combined.theta(), "reldiff", 0.05));	1✔
68
69	// Check sigma2 - using relative precision 0.05 (5%)
70	if (nk_combined.sigma2() == 0.0) {	1✔
NEW 71	CHECK(nk_updated.sigma2() == 0.0); // If combined is zero, updated should also be zero	×
72	} else {
73	CHECK(std::abs(nk_updated.sigma2() - nk_combined.sigma2()) / std::abs(nk_combined.sigma2()) < 0.05);	1✔
74	}
75
76	// Check nugget - using relative precision 0.05 (5%)
77	if (nk_combined.nugget() == 0.0) {	1✔
78	CHECK(nk_updated.nugget() == 0.0); // If combined is zero, updated should also be zero	1✔
79	} else {
NEW 80	CHECK(std::abs(nk_updated.nugget() - nk_combined.nugget()) / std::abs(nk_combined.nugget()) < 0.05);	×
81	}
82
83	// Check beta - using relative precision 0.05 (5%)
84	// Use relative comparison, which handles zero values by falling back to absolute comparison
85	CHECK(arma::approx_equal(nk_updated.beta(), nk_combined.beta(), "reldiff", 0.05));	1✔
86
87	// Test predictions on new points (main validation)
88	arma::mat X_test(5, d, arma::fill::randu);	1✔
89
90	auto pred_updated = nk_updated.predict(X_test, true, false, false);	1✔
91	auto pred_combined = nk_combined.predict(X_test, true, false, false);	1✔
92
93	INFO("Max prediction difference: " << arma::max(arma::abs(std::get<0>(pred_updated) - std::get<0>(pred_combined))));	2✔
94
95	CHECK(arma::approx_equal(std::get<0>(pred_updated), std::get<0>(pred_combined), "reldiff", 0.05));	1✔
96	}	5✔
97
98	SECTION("Multiple updates equal combined fit") {	4✔
99	// Fit model on old data
100	NuggetKriging nk_updated("gauss");	2✔
101	NuggetKriging::Parameters params{std::nullopt, true, std::nullopt, true, std::nullopt, true, std::nullopt, true};	1✔
102	nk_updated.fit(y_old, X_old, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);	1✔
103
104	// Update one point at a time
105	for (arma::uword i = 0; i < n_new; ++i) {	2✔
106	arma::vec y_single = y_new.row(i);	2✔
107	arma::mat X_single = X_new.row(i);	1✔
108	nk_updated.update(y_single, X_single, true);	1✔
109	}	1✔
110
111	// Fit model on combined data
112	NuggetKriging nk_combined("gauss");	2✔
113	nk_combined.fit(y_combined, X_combined, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);	1✔
114
115	// Test predictions
116	arma::mat X_test(5, d, arma::fill::randu);	1✔
117
118	auto pred_updated = nk_updated.predict(X_test, true, false, false);	1✔
119	auto pred_combined = nk_combined.predict(X_test, true, false, false);	1✔
120
121	INFO("Max prediction difference: " << arma::max(arma::abs(std::get<0>(pred_updated) - std::get<0>(pred_combined))));	2✔
122
123	CHECK(arma::approx_equal(std::get<0>(pred_updated), std::get<0>(pred_combined), "reldiff", 0.05));	1✔
124	}	5✔
125
126	SECTION("Different kernels") {	4✔
127	std::vector<std::string> kernels = {"gauss", "exp", "matern3_2", "matern5_2"};	7✔
128
129	for (const auto& kernel : kernels) {	5✔
130	INFO("Testing kernel: " << kernel);	4✔
131
132	// Fit and update
133	NuggetKriging nk_updated(kernel);	4✔
134	NuggetKriging::Parameters params{std::nullopt, true, std::nullopt, true, std::nullopt, true, std::nullopt, true};	4✔
135	nk_updated.fit(y_old, X_old, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);	4✔
136	nk_updated.update(y_new, X_new, true);	4✔
137
138	// Combined fit
139	NuggetKriging nk_combined(kernel);	4✔
140	nk_combined.fit(y_combined, X_combined, Trend::RegressionModel::Constant, false, "BFGS", "LL", params);	4✔
141
142	// Test predictions
143	arma::mat X_test(3, d, arma::fill::randu);	4✔
144	auto pred_updated = nk_updated.predict(X_test, true, false, false);	4✔
145	auto pred_combined = nk_combined.predict(X_test, true, false, false);	4✔
146
147	CHECK(arma::approx_equal(std::get<0>(pred_updated), std::get<0>(pred_combined), "reldiff", 0.05));	4✔
148	}	4✔
149	}	5✔
150
151	SECTION("Different trend models") {	4✔
152	std::vector<Trend::RegressionModel> trends = {
153	Trend::RegressionModel::Constant,
154	Trend::RegressionModel::Linear,
155	Trend::RegressionModel::Quadratic
156	};	1✔
157
158	for (const auto& trend : trends) {	4✔
159	INFO("Testing trend model");	3✔
160
161	// Fit and update
162	NuggetKriging nk_updated("gauss");	6✔
163	NuggetKriging::Parameters params{std::nullopt, true, std::nullopt, true, std::nullopt, true, std::nullopt, true};	3✔
164	nk_updated.fit(y_old, X_old, trend, false, "BFGS", "LL", params);	3✔
165	nk_updated.update(y_new, X_new, true);	3✔
166
167	// Combined fit
168	NuggetKriging nk_combined("gauss");	6✔
169	nk_combined.fit(y_combined, X_combined, trend, false, "BFGS", "LL", params);	3✔
170
171	// Test predictions
172	arma::mat X_test(3, d, arma::fill::randu);	3✔
173	auto pred_updated = nk_updated.predict(X_test, true, false, false);	3✔
174	auto pred_combined = nk_combined.predict(X_test, true, false, false);	3✔
175
176	CHECK(arma::approx_equal(std::get<0>(pred_updated), std::get<0>(pred_combined), "reldiff", 0.05));	3✔
177	}	3✔
178	}	5✔
179	}	4✔

libKriging / libKriging / 20211252547

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous