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

Example 1 with DataDistribution

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

the class DataUtils method boostrapFeatureDistributions.

/**
 * Generate feature distributions from an existing (evantually small) {@link DataDistribution} for each {@link Feature}.
 * Each feature intervals (min, max) and density information (mean, stdDev) are generated using bootstrap, then
 * data points are sampled from a normal distribution (see {@link #generateData(double, double, int, Random)}).
 *
 * @param dataDistribution data distribution to take feature values from
 * @param perturbationContext perturbation context
 * @param featureDistributionSize desired size of generated feature distributions
 * @param draws number of times sampling from feature values is performed
 * @param sampleSize size of each sample draw
 * @param numericFeatureZonesMap high feature score zones
 * @return a map feature name -> generated feature distribution
 */
public static Map<String, FeatureDistribution> boostrapFeatureDistributions(DataDistribution dataDistribution, PerturbationContext perturbationContext, int featureDistributionSize, int draws, int sampleSize, Map<String, HighScoreNumericFeatureZones> numericFeatureZonesMap) {
    Map<String, FeatureDistribution> featureDistributions = new HashMap<>();
    for (FeatureDistribution featureDistribution : dataDistribution.asFeatureDistributions()) {
        Feature feature = featureDistribution.getFeature();
        if (Type.NUMBER.equals(feature.getType())) {
            List<Value> values = featureDistribution.getAllSamples();
            double[] means = new double[draws];
            double[] stdDevs = new double[draws];
            double[] mins = new double[draws];
            double[] maxs = new double[draws];
            for (int i = 0; i < draws; i++) {
                List<Value> sampledValues = DataUtils.sampleWithReplacement(values, sampleSize, perturbationContext.getRandom());
                double[] data = sampledValues.stream().mapToDouble(Value::asNumber).toArray();
                double mean = DataUtils.getMean(data);
                double stdDev = Math.pow(DataUtils.getStdDev(data, mean), 2);
                double min = Arrays.stream(data).min().orElse(Double.MIN_VALUE);
                double max = Arrays.stream(data).max().orElse(Double.MAX_VALUE);
                means[i] = mean;
                stdDevs[i] = stdDev;
                mins[i] = min;
                maxs[i] = max;
            }
            double finalMean = DataUtils.getMean(means);
            double finalStdDev = Math.sqrt(DataUtils.getMean(stdDevs));
            double finalMin = DataUtils.getMean(mins);
            double finalMax = DataUtils.getMean(maxs);
            double[] doubles = DataUtils.generateData(finalMean, finalStdDev, featureDistributionSize, perturbationContext.getRandom());
            double[] boundedData = Arrays.stream(doubles).map(d -> Math.min(Math.max(d, finalMin), finalMax)).toArray();
            HighScoreNumericFeatureZones highScoreNumericFeatureZones = numericFeatureZonesMap.get(feature.getName());
            double[] finaldata;
            if (highScoreNumericFeatureZones != null) {
                double[] filteredData = DoubleStream.of(boundedData).filter(highScoreNumericFeatureZones::test).toArray();
                // only use the filtered data if it's not discarding more than 50% of the points
                if (filteredData.length > featureDistributionSize / 2) {
                    finaldata = filteredData;
                } else {
                    finaldata = boundedData;
                }
            } else {
                finaldata = boundedData;
            }
            NumericFeatureDistribution numericFeatureDistribution = new NumericFeatureDistribution(feature, finaldata);
            featureDistributions.put(feature.getName(), numericFeatureDistribution);
        }
    }
    return featureDistributions;
}

Example 2 with DataDistribution

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

the class HighScoreNumericFeatureZonesProvider method getHighScoreFeatureZones.

/**
 * Get a map of feature-name -> high score feature zones. Predictions in data distribution are sorted by (descending)
 * score, then the (aggregated) mean score is calculated and all the data points that are associated with a prediction
 * having a score between the mean and the maximum are selected (feature-wise), with an associated tolerance
 * (the stdDev of the high score feature points).
 *
 * @param dataDistribution a data distribution
 * @param predictionProvider the model used to score the inputs
 * @param features the list of features to associate high score points with
 * @param maxNoOfSamples max no. of inputs used for discovering high score zones
 * @return a map feature name -> high score numeric feature zones
 */
public static Map<String, HighScoreNumericFeatureZones> getHighScoreFeatureZones(DataDistribution dataDistribution, PredictionProvider predictionProvider, List<Feature> features, int maxNoOfSamples) {
    Map<String, HighScoreNumericFeatureZones> numericFeatureZonesMap = new HashMap<>();
    List<Prediction> scoreSortedPredictions = new ArrayList<>();
    try {
        scoreSortedPredictions.addAll(DataUtils.getScoreSortedPredictions(predictionProvider, new PredictionInputsDataDistribution(dataDistribution.sample(maxNoOfSamples))));
    } catch (ExecutionException e) {
        LOGGER.error("Could not sort predictions by score {}", e.getMessage());
    } catch (InterruptedException e) {
        LOGGER.error("Interrupted while waiting for sorting predictions by score {}", e.getMessage());
        Thread.currentThread().interrupt();
    } catch (TimeoutException e) {
        LOGGER.error("Timed out while waiting for sorting predictions by score", e);
    }
    if (!scoreSortedPredictions.isEmpty()) {
        // calculate min, max and mean scores
        double max = scoreSortedPredictions.get(0).getOutput().getOutputs().stream().mapToDouble(Output::getScore).sum();
        double min = scoreSortedPredictions.get(scoreSortedPredictions.size() - 1).getOutput().getOutputs().stream().mapToDouble(Output::getScore).sum();
        if (max != min) {
            double threshold = scoreSortedPredictions.stream().map(p -> p.getOutput().getOutputs().stream().mapToDouble(Output::getScore).sum()).mapToDouble(d -> d).average().orElse((max + min) / 2);
            // filter out predictions whose score is in [min, threshold]
            scoreSortedPredictions = scoreSortedPredictions.stream().filter(p -> p.getOutput().getOutputs().stream().mapToDouble(Output::getScore).sum() > threshold).collect(Collectors.toList());
            for (int j = 0; j < features.size(); j++) {
                Feature feature = features.get(j);
                if (Type.NUMBER.equals(feature.getType())) {
                    int finalJ = j;
                    // get feature values associated with high score inputs
                    List<Double> topValues = scoreSortedPredictions.stream().map(prediction -> prediction.getInput().getFeatures().get(finalJ).getValue().asNumber()).distinct().collect(Collectors.toList());
                    // get high score points and tolerance
                    double[] highScoreFeaturePoints = topValues.stream().flatMapToDouble(DoubleStream::of).toArray();
                    double center = DataUtils.getMean(highScoreFeaturePoints);
                    double tolerance = DataUtils.getStdDev(highScoreFeaturePoints, center) / 2;
                    HighScoreNumericFeatureZones highScoreNumericFeatureZones = new HighScoreNumericFeatureZones(highScoreFeaturePoints, tolerance);
                    numericFeatureZonesMap.put(feature.getName(), highScoreNumericFeatureZones);
                }
            }
        }
    }
    return numericFeatureZonesMap;
}

Example 3 with DataDistribution

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

the class DummyDmnModelsLimeExplainerTest method testAllTypesDMNExplanation.

@Test
void testAllTypesDMNExplanation() throws ExecutionException, InterruptedException, TimeoutException {
    DMNRuntime dmnRuntime = DMNKogito.createGenericDMNRuntime(new InputStreamReader(getClass().getResourceAsStream("/dmn/allTypes.dmn")));
    assertThat(dmnRuntime.getModels().size()).isEqualTo(1);
    final String namespace = "https://kiegroup.org/dmn/_24B9EC8C-2F02-40EB-B6BB-E8CDE82FBF08";
    final String name = "new-file";
    DecisionModel decisionModel = new DmnDecisionModel(dmnRuntime, namespace, name);
    PredictionProvider model = new DecisionModelWrapper(decisionModel);
    Map<String, Object> context = new HashMap<>();
    context.put("stringInput", "test");
    context.put("listOfStringInput", Collections.singletonList("test"));
    context.put("numberInput", 1);
    context.put("listOfNumbersInput", Collections.singletonList(1));
    context.put("booleanInput", true);
    context.put("listOfBooleansInput", Collections.singletonList(true));
    context.put("timeInput", "h09:00");
    context.put("dateInput", "2020-04-02");
    context.put("dateAndTimeInput", "2020-04-02T09:00:00");
    context.put("daysAndTimeDurationInput", "P1DT1H");
    context.put("yearsAndMonthDurationInput", "P1Y1M");
    Map<String, Object> complexInput = new HashMap<>();
    complexInput.put("aNestedListOfNumbers", Collections.singletonList(1));
    complexInput.put("aNestedString", "test");
    complexInput.put("aNestedComplexInput", Collections.singletonMap("doubleNestedNumber", 1));
    context.put("complexInput", complexInput);
    context.put("listOfComplexInput", Collections.singletonList(complexInput));
    List<Feature> features = new ArrayList<>();
    features.add(FeatureFactory.newCompositeFeature("context", context));
    PredictionInput predictionInput = new PredictionInput(features);
    List<PredictionOutput> predictionOutputs = model.predictAsync(List.of(predictionInput)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    Prediction prediction = new SimplePrediction(predictionInput, predictionOutputs.get(0));
    Random random = new Random();
    PerturbationContext perturbationContext = new PerturbationContext(0L, random, 3);
    LimeConfig limeConfig = new LimeConfig().withSamples(10).withPerturbationContext(perturbationContext);
    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()) {
        assertThat(saliency).isNotNull();
    }
    assertThatCode(() -> ValidationUtils.validateLocalSaliencyStability(model, prediction, limeExplainer, 1, 0.5, 0.2)).doesNotThrowAnyException();
    String decision = "myDecision";
    List<PredictionInput> inputs = new ArrayList<>();
    for (int n = 0; n < 10; n++) {
        inputs.add(new PredictionInput(DataUtils.perturbFeatures(features, perturbationContext)));
    }
    DataDistribution distribution = new PredictionInputsDataDistribution(inputs);
    int k = 2;
    int chunkSize = 5;
    double precision = ExplainabilityMetrics.getLocalSaliencyPrecision(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(precision).isBetween(0d, 1d);
    double recall = ExplainabilityMetrics.getLocalSaliencyRecall(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(recall).isBetween(0d, 1d);
    double f1 = ExplainabilityMetrics.getLocalSaliencyF1(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(f1).isBetween(0d, 1d);
}

Example 4 with DataDistribution

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

the class DummyDmnModelsLimeExplainerTest method testFunctional1DMNExplanation.

@Test
void testFunctional1DMNExplanation() throws ExecutionException, InterruptedException, TimeoutException {
    DMNRuntime dmnRuntime = DMNKogito.createGenericDMNRuntime(new InputStreamReader(getClass().getResourceAsStream("/dmn/functionalTest1.dmn")));
    assertThat(dmnRuntime.getModels().size()).isEqualTo(1);
    final String namespace = "https://kiegroup.org/dmn/_049CD980-1310-4B02-9E90-EFC57059F44A";
    final String name = "functionalTest1";
    DecisionModel decisionModel = new DmnDecisionModel(dmnRuntime, namespace, name);
    PredictionProvider model = new DecisionModelWrapper(decisionModel);
    Map<String, Object> context = new HashMap<>();
    context.put("booleanInput", true);
    context.put("notUsedInput", 1);
    List<Feature> features = new ArrayList<>();
    features.add(FeatureFactory.newCompositeFeature("context", context));
    PredictionInput predictionInput = new PredictionInput(features);
    List<PredictionOutput> predictionOutputs = model.predictAsync(List.of(predictionInput)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    Prediction prediction = new SimplePrediction(predictionInput, predictionOutputs.get(0));
    Random random = new Random();
    PerturbationContext perturbationContext = new PerturbationContext(0L, random, 1);
    LimeConfig limeConfig = new LimeConfig().withSamples(10).withPerturbationContext(perturbationContext);
    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()) {
        assertThat(saliency).isNotNull();
        List<FeatureImportance> topFeatures = saliency.getPositiveFeatures(2);
        assertThat(topFeatures.isEmpty()).isFalse();
        assertThat(topFeatures.get(0).getFeature().getName()).isEqualTo("booleanInput");
    }
    assertThatCode(() -> ValidationUtils.validateLocalSaliencyStability(model, prediction, limeExplainer, 1, 0.5, 0.5)).doesNotThrowAnyException();
    String decision = "decision";
    List<PredictionInput> inputs = new ArrayList<>();
    for (int n = 0; n < 10; n++) {
        inputs.add(new PredictionInput(DataUtils.perturbFeatures(features, perturbationContext)));
    }
    DataDistribution distribution = new PredictionInputsDataDistribution(inputs);
    int k = 2;
    int chunkSize = 5;
    double precision = ExplainabilityMetrics.getLocalSaliencyPrecision(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(precision).isBetween(0d, 1d);
    double recall = ExplainabilityMetrics.getLocalSaliencyRecall(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(recall).isBetween(0d, 1d);
    double f1 = ExplainabilityMetrics.getLocalSaliencyF1(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(f1).isBetween(0d, 1d);
}

Example 5 with DataDistribution

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

the class DummyDmnModelsLimeExplainerTest method testFunctional2DMNExplanation.

@Test
void testFunctional2DMNExplanation() throws ExecutionException, InterruptedException, TimeoutException {
    DMNRuntime dmnRuntime = DMNKogito.createGenericDMNRuntime(new InputStreamReader(getClass().getResourceAsStream("/dmn/functionalTest2.dmn")));
    assertThat(dmnRuntime.getModels().size()).isEqualTo(1);
    final String namespace = "https://kiegroup.org/dmn/_049CD980-1310-4B02-9E90-EFC57059F44A";
    final String name = "new-file";
    DecisionModel decisionModel = new DmnDecisionModel(dmnRuntime, namespace, name);
    PredictionProvider model = new DecisionModelWrapper(decisionModel);
    Map<String, Object> context = new HashMap<>();
    context.put("numberInput", 1);
    context.put("notUsedInput", 1);
    List<Feature> features = new ArrayList<>();
    features.add(FeatureFactory.newCompositeFeature("context", context));
    PredictionInput predictionInput = new PredictionInput(features);
    List<PredictionOutput> predictionOutputs = model.predictAsync(List.of(predictionInput)).get(Config.INSTANCE.getAsyncTimeout(), Config.INSTANCE.getAsyncTimeUnit());
    Prediction prediction = new SimplePrediction(predictionInput, predictionOutputs.get(0));
    Random random = new Random();
    PerturbationContext perturbationContext = new PerturbationContext(0L, random, 1);
    LimeConfig limeConfig = new LimeConfig().withSamples(10).withPerturbationContext(perturbationContext);
    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()) {
        assertThat(saliency).isNotNull();
        List<FeatureImportance> topFeatures = saliency.getPositiveFeatures(2);
        assertThat(topFeatures.isEmpty()).isFalse();
        assertThat(topFeatures.get(0).getFeature().getName()).isEqualTo("numberInput");
    }
    assertThatCode(() -> ValidationUtils.validateLocalSaliencyStability(model, prediction, limeExplainer, 1, 0.5, 0.5)).doesNotThrowAnyException();
    String decision = "decision";
    List<PredictionInput> inputs = new ArrayList<>();
    for (int n = 0; n < 10; n++) {
        inputs.add(new PredictionInput(DataUtils.perturbFeatures(features, perturbationContext)));
    }
    DataDistribution distribution = new PredictionInputsDataDistribution(inputs);
    int k = 2;
    int chunkSize = 5;
    double precision = ExplainabilityMetrics.getLocalSaliencyPrecision(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(precision).isBetween(0d, 1d);
    double recall = ExplainabilityMetrics.getLocalSaliencyRecall(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(recall).isBetween(0d, 1d);
    double f1 = ExplainabilityMetrics.getLocalSaliencyF1(decision, model, limeExplainer, distribution, k, chunkSize);
    assertThat(f1).isBetween(0d, 1d);
}