Examples with PerturbationContext org.kie.kogito.explainability.model.PerturbationContext used on opensource projects

Example 26 with PerturbationContext

use of org.kie.kogito.explainability.model.PerturbationContext in project kogito-apps by kiegroup.

the class PmmlCompoundScorecardLimeExplainerTest method testExplanationWeightedStabilityWithOptimization.

@Test
void testExplanationWeightedStabilityWithOptimization() throws ExecutionException, InterruptedException, TimeoutException {
    PredictionProvider model = getModel();
    List<PredictionInput> samples = getSamples();
    List<PredictionOutput> predictionOutputs = model.predictAsync(samples.subList(0, 5)).get();
    List<Prediction> predictions = DataUtils.getPredictions(samples, predictionOutputs);
    long seed = 0;
    LimeConfigOptimizer limeConfigOptimizer = new LimeConfigOptimizer().withDeterministicExecution(true).withWeightedStability(0.4, 0.6);
    Random random = new Random();
    PerturbationContext perturbationContext = new PerturbationContext(seed, random, 1);
    LimeConfig initialConfig = new LimeConfig().withSamples(10).withPerturbationContext(perturbationContext);
    LimeConfig optimizedConfig = limeConfigOptimizer.optimize(initialConfig, predictions, model);
    assertThat(optimizedConfig).isNotSameAs(initialConfig);
    LimeExplainer limeExplainer = new LimeExplainer(optimizedConfig);
    PredictionInput testPredictionInput = getTestInput();
    List<PredictionOutput> testPredictionOutputs = model.predictAsync(List.of(testPredictionInput)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    Prediction instance = new SimplePrediction(testPredictionInput, testPredictionOutputs.get(0));
    assertDoesNotThrow(() -> ValidationUtils.validateLocalSaliencyStability(model, instance, limeExplainer, 1, 0.5, 0.7));
}

Example 27 with PerturbationContext

use of org.kie.kogito.explainability.model.PerturbationContext in project kogito-apps by kiegroup.

the class PmmlRegressionCategoricalLimeExplainerTest method testExplanationStabilityWithOptimization.

@Disabled("See KOGITO-6154")
@Test
void testExplanationStabilityWithOptimization() throws ExecutionException, InterruptedException, TimeoutException {
    PredictionProvider model = getModel();
    List<PredictionInput> samples = getSamples();
    List<PredictionOutput> predictionOutputs = model.predictAsync(samples.subList(0, 5)).get();
    List<Prediction> predictions = DataUtils.getPredictions(samples, predictionOutputs);
    long seed = 0;
    LimeConfigOptimizer limeConfigOptimizer = new LimeConfigOptimizer().withDeterministicExecution(true);
    Random random = new Random();
    PerturbationContext perturbationContext = new PerturbationContext(seed, random, 1);
    LimeConfig initialConfig = new LimeConfig().withSamples(10).withPerturbationContext(perturbationContext);
    LimeConfig optimizedConfig = limeConfigOptimizer.optimize(initialConfig, predictions, model);
    assertThat(optimizedConfig).isNotSameAs(initialConfig);
    LimeExplainer limeExplainer = new LimeExplainer(optimizedConfig);
    PredictionInput testPredictionInput = getTestInput();
    List<PredictionOutput> testPredictionOutputs = model.predictAsync(List.of(testPredictionInput)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    Prediction instance = new SimplePrediction(testPredictionInput, testPredictionOutputs.get(0));
    assertDoesNotThrow(() -> ValidationUtils.validateLocalSaliencyStability(model, instance, limeExplainer, 1, 0.6, 0.6));
}

Example 28 with PerturbationContext

use of org.kie.kogito.explainability.model.PerturbationContext in project kogito-apps by kiegroup.

the class DummyModelsLimeExplainerTest method testUnusedFeatureClassification.

@ParameterizedTest
@ValueSource(longs = { 0 })
void testUnusedFeatureClassification(long seed) throws Exception {
    Random random = new Random();
    int idx = 2;
    List<Feature> features = new LinkedList<>();
    features.add(FeatureFactory.newNumericalFeature("f1", 6));
    features.add(FeatureFactory.newNumericalFeature("f2", 3));
    features.add(FeatureFactory.newNumericalFeature("f3", 5));
    PredictionProvider model = TestUtils.getEvenSumModel(idx);
    PredictionInput input = new PredictionInput(features);
    List<PredictionOutput> outputs = model.predictAsync(List.of(input)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    Prediction prediction = new SimplePrediction(input, outputs.get(0));
    LimeConfig limeConfig = new LimeConfig().withSamples(100).withPerturbationContext(new PerturbationContext(seed, random, 1));
    LimeExplainer limeExplainer = new LimeExplainer(limeConfig);
    Map<String, Saliency> saliencyMap = limeExplainer.explainAsync(prediction, model).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    for (Saliency saliency : saliencyMap.values()) {
        assertNotNull(saliency);
        List<FeatureImportance> topFeatures = saliency.getTopFeatures(3);
        assertEquals(3, topFeatures.size());
        assertEquals(1d, ExplainabilityMetrics.impactScore(model, prediction, topFeatures));
    }
    int topK = 1;
    double minimumPositiveStabilityRate = 0.5;
    double minimumNegativeStabilityRate = 0.5;
    TestUtils.assertLimeStability(model, prediction, limeExplainer, topK, minimumPositiveStabilityRate, minimumNegativeStabilityRate);
    List<PredictionInput> inputs = new ArrayList<>();
    for (int i = 0; i < 100; i++) {
        List<Feature> fs = new LinkedList<>();
        fs.add(TestUtils.getMockedNumericFeature());
        fs.add(TestUtils.getMockedNumericFeature());
        fs.add(TestUtils.getMockedNumericFeature());
        inputs.add(new PredictionInput(fs));
    }
    DataDistribution distribution = new PredictionInputsDataDistribution(inputs);
    int k = 2;
    int chunkSize = 10;
    String decision = "sum-even-but" + idx;
    double precision = ExplainabilityMetrics.getLocalSaliencyPrecision(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(precision).isEqualTo(1);
    double recall = ExplainabilityMetrics.getLocalSaliencyRecall(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(recall).isEqualTo(1);
    double f1 = ExplainabilityMetrics.getLocalSaliencyF1(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(f1).isEqualTo(1);
}

Example 29 with PerturbationContext

use of org.kie.kogito.explainability.model.PerturbationContext in project kogito-apps by kiegroup.

the class DummyModelsLimeExplainerTest method testTextSpamClassification.

@ParameterizedTest
@ValueSource(longs = { 0 })
void testTextSpamClassification(long seed) throws Exception {
    Random random = new Random();
    List<Feature> features = new LinkedList<>();
    Function<String, List<String>> tokenizer = s -> Arrays.asList(s.split(" ").clone());
    features.add(FeatureFactory.newFulltextFeature("f1", "we go here and there", tokenizer));
    features.add(FeatureFactory.newFulltextFeature("f2", "please give me some money", tokenizer));
    features.add(FeatureFactory.newFulltextFeature("f3", "dear friend, please reply", tokenizer));
    PredictionInput input = new PredictionInput(features);
    PredictionProvider model = TestUtils.getDummyTextClassifier();
    List<PredictionOutput> outputs = model.predictAsync(List.of(input)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    Prediction prediction = new SimplePrediction(input, outputs.get(0));
    LimeConfig limeConfig = new LimeConfig().withSamples(100).withPerturbationContext(new PerturbationContext(seed, random, 1));
    LimeExplainer limeExplainer = new LimeExplainer(limeConfig);
    Map<String, Saliency> saliencyMap = limeExplainer.explainAsync(prediction, model).toCompletableFuture().get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    for (Saliency saliency : saliencyMap.values()) {
        assertNotNull(saliency);
        List<FeatureImportance> topFeatures = saliency.getPositiveFeatures(1);
        assertEquals(1, topFeatures.size());
        assertEquals(1d, ExplainabilityMetrics.impactScore(model, prediction, topFeatures));
    }
    int topK = 1;
    double minimumPositiveStabilityRate = 0.5;
    double minimumNegativeStabilityRate = 0.2;
    TestUtils.assertLimeStability(model, prediction, limeExplainer, topK, minimumPositiveStabilityRate, minimumNegativeStabilityRate);
    List<PredictionInput> inputs = new ArrayList<>();
    for (int i = 0; i < 100; i++) {
        List<Feature> fs = new LinkedList<>();
        fs.add(TestUtils.getMockedNumericFeature());
        fs.add(TestUtils.getMockedNumericFeature());
        fs.add(TestUtils.getMockedNumericFeature());
        inputs.add(new PredictionInput(fs));
    }
    DataDistribution distribution = new PredictionInputsDataDistribution(inputs);
    int k = 2;
    int chunkSize = 10;
    String decision = "spam";
    double precision = ExplainabilityMetrics.getLocalSaliencyPrecision(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(precision).isEqualTo(1);
    double recall = ExplainabilityMetrics.getLocalSaliencyRecall(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(recall).isEqualTo(1);
    double f1 = ExplainabilityMetrics.getLocalSaliencyF1(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(f1).isEqualTo(1);
}

Example 30 with PerturbationContext

use of org.kie.kogito.explainability.model.PerturbationContext in project kogito-apps by kiegroup.

the class RecordingLimeExplainerTest method testAutomaticConfigOptimization.

@ParameterizedTest
@ValueSource(longs = { 0 })
void testAutomaticConfigOptimization(long seed) throws Exception {
    PredictionProvider model = TestUtils.getSumThresholdModel(10, 10);
    PerturbationContext pc = new PerturbationContext(seed, new Random(), 1);
    LimeConfig config = new LimeConfig().withPerturbationContext(pc);
    RecordingLimeExplainer limeExplainer = new RecordingLimeExplainer(2);
    for (int i = 0; i < 50; i++) {
        List<Feature> features = new LinkedList<>();
        features.add(TestUtils.getMockedNumericFeature(Type.NUMBER.randomValue(pc).asNumber()));
        features.add(TestUtils.getMockedNumericFeature(Type.NUMBER.randomValue(pc).asNumber()));
        features.add(TestUtils.getMockedNumericFeature(Type.NUMBER.randomValue(pc).asNumber()));
        PredictionInput input = new PredictionInput(features);
        List<PredictionOutput> outputs = model.predictAsync(List.of(input)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
        Prediction prediction = new SimplePrediction(input, outputs.get(0));
        Map<String, Saliency> saliencyMap = limeExplainer.explainAsync(prediction, model).toCompletableFuture().get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
        for (Saliency saliency : saliencyMap.values()) {
            assertNotNull(saliency);
        }
    }
    LimeConfig optimizedConfig = limeExplainer.getExecutionConfig();
    assertThat(optimizedConfig).isNotEqualTo(config);
}