Examples with ModelsComposition org.apache.ignite.ml.composition.ModelsComposition used on opensource projects

Example 11 with ModelsComposition

use of org.apache.ignite.ml.composition.ModelsComposition in project ignite by apache.

the class GDBOnTreesLearningStrategy method update.

/**
 * {@inheritDoc}
 */
@Override
public <K, V> List<IgniteModel<Vector, Double>> update(GDBModel mdlToUpdate, DatasetBuilder<K, V> datasetBuilder, Preprocessor<K, V> vectorizer) {
    LearningEnvironment environment = envBuilder.buildForTrainer();
    environment.initDeployingContext(vectorizer);
    DatasetTrainer<? extends IgniteModel<Vector, Double>, Double> trainer = baseMdlTrainerBuilder.get();
    assert trainer instanceof DecisionTreeTrainer;
    DecisionTreeTrainer decisionTreeTrainer = (DecisionTreeTrainer) trainer;
    List<IgniteModel<Vector, Double>> models = initLearningState(mdlToUpdate);
    ConvergenceChecker<K, V> convCheck = checkConvergenceStgyFactory.create(sampleSize, externalLbToInternalMapping, loss, datasetBuilder, vectorizer);
    try (Dataset<EmptyContext, DecisionTreeData> dataset = datasetBuilder.build(envBuilder, new EmptyContextBuilder<>(), new DecisionTreeDataBuilder<>(vectorizer, useIdx), environment)) {
        for (int i = 0; i < cntOfIterations; i++) {
            double[] weights = Arrays.copyOf(compositionWeights, models.size());
            WeightedPredictionsAggregator aggregator = new WeightedPredictionsAggregator(weights, meanLbVal);
            ModelsComposition currComposition = new ModelsComposition(models, aggregator);
            if (convCheck.isConverged(dataset, currComposition))
                break;
            dataset.compute(part -> {
                if (part.getCopiedOriginalLabels() == null)
                    part.setCopiedOriginalLabels(Arrays.copyOf(part.getLabels(), part.getLabels().length));
                for (int j = 0; j < part.getLabels().length; j++) {
                    double mdlAnswer = currComposition.predict(VectorUtils.of(part.getFeatures()[j]));
                    double originalLbVal = externalLbToInternalMapping.apply(part.getCopiedOriginalLabels()[j]);
                    part.getLabels()[j] = -loss.gradient(sampleSize, originalLbVal, mdlAnswer);
                }
            });
            long startTs = System.currentTimeMillis();
            models.add(decisionTreeTrainer.fit(dataset));
            double learningTime = (double) (System.currentTimeMillis() - startTs) / 1000.0;
            trainerEnvironment.logger(getClass()).log(MLLogger.VerboseLevel.LOW, "One model training time was %.2fs", learningTime);
        }
    } catch (Exception e) {
        throw new RuntimeException(e);
    }
    compositionWeights = Arrays.copyOf(compositionWeights, models.size());
    return models;
}

Example 12 with ModelsComposition

use of org.apache.ignite.ml.composition.ModelsComposition in project ignite by apache.

the class GDBTrainerTest method testFitRegression.

/**
 */
@Test
public void testFitRegression() {
    int size = 100;
    double[] xs = new double[size];
    double[] ys = new double[size];
    double from = -5.0;
    double to = 5.0;
    double step = Math.abs(from - to) / size;
    Map<Integer, double[]> learningSample = new HashMap<>();
    for (int i = 0; i < size; i++) {
        xs[i] = from + step * i;
        ys[i] = 2 * xs[i];
        learningSample.put(i, new double[] { xs[i], ys[i] });
    }
    GDBTrainer trainer = new GDBRegressionOnTreesTrainer(1.0, 2000, 3, 0.0).withUsingIdx(true);
    IgniteModel<Vector, Double> mdl = trainer.fit(learningSample, 1, new DoubleArrayVectorizer<Integer>().labeled(Vectorizer.LabelCoordinate.LAST));
    double mse = 0.0;
    for (int j = 0; j < size; j++) {
        double x = xs[j];
        double y = ys[j];
        double p = mdl.predict(VectorUtils.of(x));
        mse += Math.pow(y - p, 2);
    }
    mse /= size;
    assertEquals(0.0, mse, 0.0001);
    ModelsComposition composition = (ModelsComposition) mdl;
    assertTrue(!composition.toString().isEmpty());
    assertTrue(!composition.toString(true).isEmpty());
    assertTrue(!composition.toString(false).isEmpty());
    composition.getModels().forEach(m -> assertTrue(m instanceof DecisionTreeModel));
    assertEquals(2000, composition.getModels().size());
    assertTrue(composition.getPredictionsAggregator() instanceof WeightedPredictionsAggregator);
    trainer = trainer.withCheckConvergenceStgyFactory(new MeanAbsValueConvergenceCheckerFactory(0.1));
    assertTrue(trainer.fit(learningSample, 1, new DoubleArrayVectorizer<Integer>().labeled(1)).getModels().size() < 2000);
}