[ML] Ensure estimated latitude is within the allowed range (#2586)

Fixes #2578

Author: valeriy42

docs/CHANGELOG.asciidoc

@@ -34,6 +34,10 @@
 
 * Upgrade Boost libraries to version 1.83. (See {ml-pull}2560[#2560].)
 
+=== Bug Fixes
+
+* Ensure the estimated latitude is within the allowed range (See {ml-pull}#2586[2586].)
+
 == {es} version 8.11.0
 
 === Enhancements

include/model/CIndividualModel.h

@@ -28,6 +28,7 @@
 namespace ml {
 namespace model {
 class CAnnotatedProbabilityBuilder;
+class CIndividualModelTestHelper;
 class CProbabilityAndInfluenceCalculator;
 
 //! \brief The most basic individual model interface.
@@ -303,6 +304,8 @@ class MODEL_EXPORT CIndividualModel : public CAnomalyDetectorModel {
 
     //! The memory estimator.
     mutable CMemoryUsageEstimator m_MemoryEstimator;
+
+    friend class CIndividualModelTestHelper;
 };
 }
 }

lib/model/ModelTypes.cc

@@ -902,7 +902,7 @@ TDouble1VecDouble1VecPr support(EFeature feature) {
     case E_IndividualHighVarianceByPerson:
         return {TDouble1Vec(d, 0.0), TDouble1Vec(d, MAX_DOUBLE)};
     case E_IndividualMeanLatLongByPerson:
-        return {TDouble1Vec(d, -180.0), TDouble1Vec(d, 180.0)};
+        return {TDouble1Vec{-90, -180.0}, TDouble1Vec{90, 180.0}};
 
     case E_PopulationAttributeTotalCountByPerson:
     case E_PopulationCountByBucketPersonAndAttribute:
@@ -945,7 +945,7 @@ TDouble1VecDouble1VecPr support(EFeature feature) {
     case E_PopulationHighVarianceByPersonAndAttribute:
         return {TDouble1Vec(d, 0.0), TDouble1Vec(d, MAX_DOUBLE)};
     case E_PopulationMeanLatLongByPersonAndAttribute:
-        return {TDouble1Vec(d, -180.0), TDouble1Vec(d, 180.0)};
+        return {TDouble1Vec{-90, -180.0}, TDouble1Vec{90, 180.0}};
     }
 
     return {TDouble1Vec(d, MIN_DOUBLE), TDouble1Vec(d, MAX_DOUBLE)};

lib/model/unittest/CMetricModelTest.cc

@@ -27,6 +27,7 @@
 #include <model/CDataGatherer.h>
 #include <model/CDetectionRule.h>
 #include <model/CEventData.h>
+#include <model/CIndividualModel.h>
 #include <model/CInterimBucketCorrector.h>
 #include <model/CMetricModel.h>
 #include <model/CMetricModelFactory.h>
@@ -50,6 +51,18 @@
 #include <utility>
 #include <vector>
 
+namespace ml {
+namespace model {
+class CIndividualModelTestHelper {
+public:
+    static void setFeature(ml::model::CIndividualModel& model) {
+        auto& feature = model.m_FeatureModels[0];
+        feature.s_Models.emplace_back(feature.s_NewModel->clone(0));
+    }
+};
+}
+}
+
 BOOST_AUTO_TEST_SUITE(CMetricModelTest)
 
 using namespace ml;
@@ -2349,4 +2362,77 @@ BOOST_FIXTURE_TEST_CASE(testIgnoreSamplingGivenDetectionRules, CTestFixture) {
                         modelNoSkipTime);
 }
 
+class MyFakeModel : public ml::maths::common::CModelStub {
+public:
+    MyFakeModel(model_t::TDouble2Vec latLong) : m_LatLong(latLong) {}
+
+    model_t::TDouble2Vec predict(core_t::TTime /*time*/,
+                                 const TSizeDoublePr1Vec& /*correlated*/,
+                                 TDouble2Vec /*hint*/) const override {
+        return m_LatLong;
+    }
+
+    CModelStub* clone(std::size_t /*id*/) const override {
+        return new MyFakeModel(m_LatLong);
+    }
+
+private:
+    model_t::TDouble2Vec m_LatLong;
+};
+
+BOOST_FIXTURE_TEST_CASE(testLatLongNotMalformed, CTestFixture) {
+    // This test ensures that the latitudes and longitudes generated by the model are within the
+    // expected range.
+
+    // initialize the model
+    core_t::TTime startTime{45};
+    core_t::TTime bucketLength{5};
+    model_t::TFeatureVec features{model_t::E_IndividualMeanLatLongByPerson};
+    SModelParams params(bucketLength);
+    params.s_InitialDecayRateMultiplier = 1.0;
+    params.s_MaximumUpdatesPerBucket = 0.0;
+    size_t sampleCount{1};
+
+    this->makeModel(params, features, startTime, sampleCount);
+
+    ml::model::CAnomalyDetectorModel::TFeatureMultivariatePriorSPtrPrVec newFeatureCorelateModelPriors;
+    ml::model::CAnomalyDetectorModel::TFeatureCorrelationsPtrPrVec featureCorrelatesModels;
+    ml::model::CAnomalyDetectorModel::TFeatureInfluenceCalculatorCPtrPrVecVec influenceCalculators;
+
+    // generate random numbers for latitudes and longitudes in the range [-360, 360]
+    test::CRandomNumbers rng;
+    int numberOfTrials{100};
+    std::vector<double> latitudes;
+    std::vector<double> longitudes;
+    rng.generateUniformSamples(-360.0, 360.0, numberOfTrials, latitudes);
+    rng.generateUniformSamples(-360.0, 360.0, numberOfTrials, longitudes);
+
+    for (auto i = 0; i < numberOfTrials; ++i) {
+        ml::model::CAnomalyDetectorModel::TFeatureMathsModelSPtrPrVec newFeatureModels = {
+            std::make_pair(model_t::E_IndividualMeanLatLongByPerson,
+                           std::make_shared<MyFakeModel>(model_t::TDouble2Vec(
+                               {latitudes[i], longitudes[i]})))};
+        ml::model::CMetricModel model{params,
+                                      m_Gatherer,
+                                      newFeatureModels,
+                                      newFeatureCorelateModelPriors,
+                                      std::move(featureCorrelatesModels),
+                                      influenceCalculators,
+                                      m_InterimBucketCorrector};
+        CIndividualModelTestHelper::setFeature(model);
+        model_t::CResultType type(model_t::CResultType::E_Unconditional |
+                                  model_t::CResultType::E_Final);
+        core_t::TTime time{startTime};
+        auto result = model.baselineBucketMean(features[0], 0, 0, type, NO_CORRELATES, time);
+        // ensure the result is withing the expected range
+        BOOST_REQUIRE_EQUAL(result.size(), 2);
+        // Check latitude in [-90, 90]
+        BOOST_TEST_REQUIRE(result[0] >= -90.0);
+        BOOST_TEST_REQUIRE(result[0] <= 90.0);
+        // Check longitude in [-180, 180]
+        BOOST_TEST_REQUIRE(result[1] >= -180.0);
+        BOOST_TEST_REQUIRE(result[1] <= 180.0);
+    }
+}
+
 BOOST_AUTO_TEST_SUITE_END()