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

Example 6 with PartialDependenceGraph

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

the class PartialDependencePlotExplainerTest method testTextClassifier.

@ParameterizedTest
@ValueSource(ints = { 0, 1, 2, 3, 4 })
void testTextClassifier(int seed) throws Exception {
    Random random = new Random();
    random.setSeed(seed);
    PartialDependencePlotExplainer partialDependencePlotExplainer = new PartialDependencePlotExplainer();
    PredictionProvider model = TestUtils.getDummyTextClassifier();
    Collection<Prediction> predictions = new ArrayList<>(3);
    List<String> texts = List.of("we want your money", "please reply quickly", "you are the lucky winner", "huge donation for you!", "bitcoin for you");
    for (String text : texts) {
        List<Feature> features = new ArrayList<>();
        features.add(FeatureFactory.newFulltextFeature("text", text));
        PredictionInput predictionInput = new PredictionInput(features);
        PredictionOutput predictionOutput = model.predictAsync(List.of(predictionInput)).get().get(0);
        predictions.add(new SimplePrediction(predictionInput, predictionOutput));
    }
    List<PartialDependenceGraph> pdps = partialDependencePlotExplainer.explainFromPredictions(model, predictions);
    assertThat(pdps).isNotEmpty();
}

Example 7 with PartialDependenceGraph

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

the class PartialDependencePlotExplainer method getPartialDependenceGraph.

private PartialDependenceGraph getPartialDependenceGraph(PredictionProvider model, List<PredictionInput> trainingData, List<Value> xsValues, List<Feature> featureXSvalues, int outputIndex) throws InterruptedException, ExecutionException, TimeoutException {
    Output outputDecision = null;
    Feature feature = null;
    // each feature value of the feature under analysis should have a corresponding output value (composed by the marginal impacts of the other features)
    List<Map<Value, Long>> valueCounts = new ArrayList<>(featureXSvalues.size());
    for (int i = 0; i < featureXSvalues.size(); i++) {
        // initialize an empty feature to use in the generated PDP
        if (feature == null) {
            feature = FeatureFactory.copyOf(featureXSvalues.get(i), new Value(null));
        }
        List<PredictionInput> predictionInputs = prepareInputs(featureXSvalues.get(i), trainingData);
        List<PredictionOutput> predictionOutputs = getOutputs(model, predictionInputs);
        // prediction requests are batched per value of feature 'Xs' under analysis
        for (PredictionOutput predictionOutput : predictionOutputs) {
            Output output = predictionOutput.getOutputs().get(outputIndex);
            if (outputDecision == null) {
                outputDecision = new Output(output.getName(), output.getType());
            }
            // update output value counts
            updateValueCounts(valueCounts, i, output);
        }
    }
    if (outputDecision != null) {
        List<Value> yValues = collapseMarginalImpacts(valueCounts, outputDecision.getType());
        return new PartialDependenceGraph(feature, outputDecision, xsValues, yValues);
    } else {
        throw new IllegalArgumentException("cannot produce PDP for null decision");
    }
}

Example 8 with PartialDependenceGraph

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

the class PartialDependencePlotExplainer method explainFromDataDistribution.

private List<PartialDependenceGraph> explainFromDataDistribution(PredictionProvider model, int outputSize, DataDistribution dataDistribution) throws InterruptedException, ExecutionException, TimeoutException {
    long start = System.currentTimeMillis();
    List<PartialDependenceGraph> pdps = new ArrayList<>();
    List<FeatureDistribution> featureDistributions = dataDistribution.asFeatureDistributions();
    // fetch entire data distributions for all features
    List<PredictionInput> trainingData = dataDistribution.sample(config.getSeriesLength());
    // create a PDP for each feature
    for (FeatureDistribution featureDistribution : featureDistributions) {
        // generate (further) samples for the feature under analysis
        // TBD: maybe just reuse trainingData
        List<Value> xsValues = featureDistribution.sample(config.getSeriesLength()).stream().sorted(// sort alphanumerically (if Value#asNumber is NaN)
        Comparator.comparing(Value::asString)).sorted(// sort by natural order
        (v1, v2) -> Comparator.comparingDouble(Value::asNumber).compare(v1, v2)).distinct().collect(Collectors.toList());
        List<Feature> featureXSvalues = // transform sampled Values into Features
        xsValues.stream().map(v -> FeatureFactory.copyOf(featureDistribution.getFeature(), v)).collect(Collectors.toList());
        // create a PDP for each feature and each output
        for (int outputIndex = 0; outputIndex < outputSize; outputIndex++) {
            PartialDependenceGraph partialDependenceGraph = getPartialDependenceGraph(model, trainingData, xsValues, featureXSvalues, outputIndex);
            pdps.add(partialDependenceGraph);
        }
    }
    long end = System.currentTimeMillis();
    LOGGER.debug("explanation time: {}ms", (end - start));
    return pdps;
}

Example 9 with PartialDependenceGraph

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

the class TrafficViolationDmnPDPExplainerTest method testTrafficViolationDMNExplanation.

@Test
void testTrafficViolationDMNExplanation() throws ExecutionException, InterruptedException, TimeoutException {
    DMNRuntime dmnRuntime = DMNKogito.createGenericDMNRuntime(new InputStreamReader(getClass().getResourceAsStream("/dmn/TrafficViolation.dmn")));
    assertEquals(1, dmnRuntime.getModels().size());
    final String TRAFFIC_VIOLATION_NS = "https://github.com/kiegroup/drools/kie-dmn/_A4BCA8B8-CF08-433F-93B2-A2598F19ECFF";
    final String TRAFFIC_VIOLATION_NAME = "Traffic Violation";
    DecisionModel decisionModel = new DmnDecisionModel(dmnRuntime, TRAFFIC_VIOLATION_NS, TRAFFIC_VIOLATION_NAME);
    PredictionProvider model = new DecisionModelWrapper(decisionModel);
    List<PredictionInput> inputs = DmnTestUtils.randomTrafficViolationInputs();
    List<PredictionOutput> predictionOutputs = model.predictAsync(inputs).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    List<Prediction> predictions = new ArrayList<>();
    for (int i = 0; i < predictionOutputs.size(); i++) {
        predictions.add(new SimplePrediction(inputs.get(i), predictionOutputs.get(i)));
    }
    PartialDependencePlotExplainer partialDependencePlotExplainer = new PartialDependencePlotExplainer();
    List<PartialDependenceGraph> pdps = partialDependencePlotExplainer.explainFromPredictions(model, predictions);
    AssertionsForClassTypes.assertThat(pdps).isNotNull();
    Assertions.assertThat(pdps).hasSize(8);
}

Example 10 with PartialDependenceGraph

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

the class PrequalificationDmnPDPExplainerTest method testPrequalificationDMNExplanation.

@Test
void testPrequalificationDMNExplanation() throws ExecutionException, InterruptedException, TimeoutException {
    DMNRuntime dmnRuntime = DMNKogito.createGenericDMNRuntime(new InputStreamReader(getClass().getResourceAsStream("/dmn/Prequalification-1.dmn")));
    assertEquals(1, dmnRuntime.getModels().size());
    final String NS = "http://www.trisotech.com/definitions/_f31e1f8e-d4ce-4a3a-ac3b-747efa6b3401";
    final String NAME = "Prequalification";
    DecisionModel decisionModel = new DmnDecisionModel(dmnRuntime, NS, NAME);
    PredictionProvider model = new DecisionModelWrapper(decisionModel);
    List<PredictionInput> inputs = DmnTestUtils.randomPrequalificationInputs();
    List<PredictionOutput> predictionOutputs = model.predictAsync(inputs).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    List<Prediction> predictions = new ArrayList<>();
    for (int i = 0; i < predictionOutputs.size(); i++) {
        predictions.add(new SimplePrediction(inputs.get(i), predictionOutputs.get(i)));
    }
    PartialDependencePlotExplainer partialDependencePlotExplainer = new PartialDependencePlotExplainer();
    List<PartialDependenceGraph> pdps = partialDependencePlotExplainer.explainFromPredictions(model, predictions);
    AssertionsForClassTypes.assertThat(pdps).isNotNull();
    Assertions.assertThat(pdps).hasSize(25);
}