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

Example 6 with CounterfactualPrediction

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

the class CounterfactualExplainerServiceHandlerTest method testGetPredictionWithFlatOutputModelReordered.

@Test
public void testGetPredictionWithFlatOutputModelReordered() {
    CounterfactualExplainabilityRequest request = new CounterfactualExplainabilityRequest(EXECUTION_ID, SERVICE_URL, MODEL_IDENTIFIER, COUNTERFACTUAL_ID, Collections.emptyList(), List.of(new NamedTypedValue("inputsAreValid", new UnitValue("boolean", BooleanNode.FALSE)), new NamedTypedValue("canRequestLoan", new UnitValue("booelan", BooleanNode.TRUE)), new NamedTypedValue("my-scoring-function", new UnitValue("number", new DoubleNode(0.85)))), Collections.emptyList(), MAX_RUNNING_TIME_SECONDS);
    Prediction prediction = handler.getPrediction(request);
    assertTrue(prediction instanceof CounterfactualPrediction);
    CounterfactualPrediction counterfactualPrediction = (CounterfactualPrediction) prediction;
    List<Output> outputs = counterfactualPrediction.getOutput().getOutputs();
    assertEquals(3, outputs.size());
    Output output1 = outputs.get(0);
    assertEquals("my-scoring-function", output1.getName());
    assertEquals(Type.NUMBER, output1.getType());
    assertEquals(0.85, output1.getValue().asNumber());
    Output output2 = outputs.get(1);
    assertEquals("inputsAreValid", output2.getName());
    assertEquals(Type.BOOLEAN, output2.getType());
    assertEquals(Boolean.FALSE, output2.getValue().getUnderlyingObject());
    Output output3 = outputs.get(2);
    assertEquals("canRequestLoan", output3.getName());
    assertEquals(Type.BOOLEAN, output3.getType());
    assertEquals(Boolean.TRUE, output3.getValue().getUnderlyingObject());
    assertTrue(counterfactualPrediction.getInput().getFeatures().isEmpty());
    assertEquals(counterfactualPrediction.getMaxRunningTimeSeconds(), request.getMaxRunningTimeSeconds());
}

Example 7 with CounterfactualPrediction

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

the class CounterfactualExplainerServiceHandler method getPrediction.

@Override
public Prediction getPrediction(CounterfactualExplainabilityRequest request) {
    Collection<NamedTypedValue> goals = toMapBasedSorting(request.getGoals());
    Collection<CounterfactualSearchDomain> searchDomains = request.getSearchDomains();
    Collection<NamedTypedValue> originalInputs = request.getOriginalInputs();
    Long maxRunningTimeSeconds = request.getMaxRunningTimeSeconds();
    if (Objects.nonNull(maxRunningTimeSeconds)) {
        if (maxRunningTimeSeconds > kafkaMaxRecordAgeSeconds) {
            LOGGER.info(String.format("Maximum Running Timeout set to '%d's since the provided value '%d's exceeded the Messaging sub-system configuration '%d's.", kafkaMaxRecordAgeSeconds, maxRunningTimeSeconds, kafkaMaxRecordAgeSeconds));
            maxRunningTimeSeconds = kafkaMaxRecordAgeSeconds;
        }
    }
    // See https://issues.redhat.com/browse/FAI-473 and https://issues.redhat.com/browse/FAI-474
    if (isUnsupportedModel(originalInputs, goals, searchDomains)) {
        throw new IllegalArgumentException("Counterfactual explanations only support flat models.");
    }
    PredictionInput input = new PredictionInput(toFeatureList(originalInputs, searchDomains));
    PredictionOutput output = new PredictionOutput(toOutputList(goals));
    return new CounterfactualPrediction(input, output, null, UUID.fromString(request.getExecutionId()), maxRunningTimeSeconds);
}

Example 8 with CounterfactualPrediction

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

the class LoanEligibilityDmnCounterfactualExplainerTest method testLoanEligibilityDMNExplanation.

@Test
void testLoanEligibilityDMNExplanation() throws ExecutionException, InterruptedException, TimeoutException {
    PredictionProvider model = getModel();
    final List<Output> goal = List.of(new Output("Is Enought?", Type.NUMBER, new Value(100), 0.0d), new Output("Eligibility", Type.TEXT, new Value("No"), 0.0d), new Output("Decide", Type.BOOLEAN, new Value(true), 0.0d));
    final TerminationConfig terminationConfig = new TerminationConfig().withScoreCalculationCountLimit(steps);
    final SolverConfig solverConfig = SolverConfigBuilder.builder().withTerminationConfig(terminationConfig).build();
    solverConfig.setRandomSeed(randomSeed);
    solverConfig.setEnvironmentMode(EnvironmentMode.REPRODUCIBLE);
    CounterfactualConfig config = new CounterfactualConfig();
    config.withSolverConfig(solverConfig);
    final CounterfactualExplainer explainer = new CounterfactualExplainer(config);
    PredictionInput input = getTestInput();
    PredictionOutput output = new PredictionOutput(goal);
    // test model
    List<PredictionOutput> predictionOutputs = model.predictAsync(List.of(input)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    Prediction prediction = new CounterfactualPrediction(input, output, null, UUID.randomUUID(), null);
    CounterfactualResult counterfactualResult = explainer.explainAsync(prediction, model).get();
    List<Feature> cfFeatures = counterfactualResult.getEntities().stream().map(CounterfactualEntity::asFeature).collect(Collectors.toList());
    List<Feature> unflattened = CompositeFeatureUtils.unflattenFeatures(cfFeatures, input.getFeatures());
    List<PredictionOutput> outputs = model.predictAsync(List.of(new PredictionInput(unflattened))).get();
    assertTrue(counterfactualResult.isValid());
    final Output decideOutput = outputs.get(0).getOutputs().get(2);
    assertEquals("Decide", decideOutput.getName());
    assertTrue((Boolean) decideOutput.getValue().getUnderlyingObject());
}

Example 9 with CounterfactualPrediction

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

the class ComplexEligibilityDmnCounterfactualExplainerTest method testDMNScoringFunction.

@Test
void testDMNScoringFunction() throws ExecutionException, InterruptedException, TimeoutException {
    PredictionProvider model = getModel();
    final List<Output> goal = generateGoal(true, true, 1.0);
    List<Feature> features = new LinkedList<>();
    features.add(FeatureFactory.newNumericalFeature("age", 40, NumericalFeatureDomain.create(18, 60)));
    features.add(FeatureFactory.newBooleanFeature("hasReferral", true));
    features.add(FeatureFactory.newNumericalFeature("monthlySalary", 500, NumericalFeatureDomain.create(10, 100_000)));
    final TerminationConfig terminationConfig = new TerminationConfig().withScoreCalculationCountLimit(10_000L);
    // for the purpose of this test, only a few steps are necessary
    final SolverConfig solverConfig = SolverConfigBuilder.builder().withTerminationConfig(terminationConfig).build();
    solverConfig.setRandomSeed((long) 23);
    solverConfig.setEnvironmentMode(EnvironmentMode.REPRODUCIBLE);
    final CounterfactualConfig counterfactualConfig = new CounterfactualConfig().withSolverConfig(solverConfig).withGoalThreshold(0.01);
    final CounterfactualExplainer counterfactualExplainer = new CounterfactualExplainer(counterfactualConfig);
    PredictionInput input = new PredictionInput(features);
    PredictionOutput output = new PredictionOutput(goal);
    Prediction prediction = new CounterfactualPrediction(input, output, null, UUID.randomUUID(), 60L);
    final CounterfactualResult counterfactualResult = counterfactualExplainer.explainAsync(prediction, model).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    List<Output> cfOutputs = counterfactualResult.getOutput().get(0).getOutputs();
    assertTrue(counterfactualResult.isValid());
    assertEquals("inputsAreValid", cfOutputs.get(0).getName());
    assertTrue((Boolean) cfOutputs.get(0).getValue().getUnderlyingObject());
    assertEquals("canRequestLoan", cfOutputs.get(1).getName());
    assertTrue((Boolean) cfOutputs.get(1).getValue().getUnderlyingObject());
    assertEquals("my-scoring-function", cfOutputs.get(2).getName());
    assertEquals(1.0, ((BigDecimal) cfOutputs.get(2).getValue().getUnderlyingObject()).doubleValue(), 0.01);
    List<CounterfactualEntity> entities = counterfactualResult.getEntities();
    assertEquals("age", entities.get(0).asFeature().getName());
    assertEquals(18, entities.get(0).asFeature().getValue().asNumber());
    assertEquals("hasReferral", entities.get(1).asFeature().getName());
    assertTrue((Boolean) entities.get(1).asFeature().getValue().getUnderlyingObject());
    assertEquals("monthlySalary", entities.get(2).asFeature().getName());
    final double monthlySalary = entities.get(2).asFeature().getValue().asNumber();
    assertEquals(7900, monthlySalary, 10);
    // since the scoring function is ((0.6 * ((42 - age + 18)/42)) + (0.4 * (monthlySalary/8000)))
    // for a result of 1.0 the relation must be age = (7*monthlySalary)/2000 - 10
    assertEquals(18, (7 * monthlySalary) / 2000.0 - 10.0, 0.5);
}

Example 10 with CounterfactualPrediction

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

the class ComplexEligibilityDmnCounterfactualExplainerTest method testDMNInvalidCounterfactualExplanation.

@Test
void testDMNInvalidCounterfactualExplanation() throws ExecutionException, InterruptedException, TimeoutException {
    PredictionProvider model = getModel();
    final List<Output> goal = generateGoal(true, true, 0.6);
    List<Feature> features = new LinkedList<>();
    // DMN model does not allow loans for age >= 60, so no CF will be possible
    features.add(FeatureFactory.newNumericalFeature("age", 61));
    features.add(FeatureFactory.newBooleanFeature("hasReferral", true));
    features.add(FeatureFactory.newNumericalFeature("monthlySalary", 500, NumericalFeatureDomain.create(10, 10_000)));
    final TerminationConfig terminationConfig = new TerminationConfig().withScoreCalculationCountLimit(10_000L);
    // for the purpose of this test, only a few steps are necessary
    final SolverConfig solverConfig = SolverConfigBuilder.builder().withTerminationConfig(terminationConfig).build();
    solverConfig.setRandomSeed((long) 23);
    solverConfig.setEnvironmentMode(EnvironmentMode.REPRODUCIBLE);
    final CounterfactualConfig counterfactualConfig = new CounterfactualConfig().withSolverConfig(solverConfig).withGoalThreshold(0.01);
    final CounterfactualExplainer counterfactualExplainer = new CounterfactualExplainer(counterfactualConfig);
    PredictionInput input = new PredictionInput(features);
    PredictionOutput output = new PredictionOutput(goal);
    Prediction prediction = new CounterfactualPrediction(input, output, null, UUID.randomUUID(), 60L);
    final CounterfactualResult counterfactualResult = counterfactualExplainer.explainAsync(prediction, model).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    assertFalse(counterfactualResult.isValid());
}