org.jpmml.converter.CMatrixUtil Java Examples

@Override
public MiningModel encodeMiningModel(List<RegTree> trees, List<Float> weights, float base_score, Integer ntreeLimit, Schema schema){
	Schema segmentSchema = schema.toAnonymousRegressorSchema(DataType.FLOAT);

	List<MiningModel> miningModels = new ArrayList<>();

	CategoricalLabel categoricalLabel = (CategoricalLabel)schema.getLabel();

	for(int i = 0, columns = categoricalLabel.size(), rows = (trees.size() / columns); i < columns; i++){
		MiningModel miningModel = createMiningModel(CMatrixUtil.getColumn(trees, rows, columns, i), (weights != null) ? CMatrixUtil.getColumn(weights, rows, columns, i) : null, base_score, ntreeLimit, segmentSchema)
			.setOutput(ModelUtil.createPredictedOutput(FieldName.create("xgbValue(" + categoricalLabel.getValue(i) + ")"), OpType.CONTINUOUS, DataType.FLOAT));

		miningModels.add(miningModel);
	}

	return MiningModelUtil.createClassification(miningModels, RegressionModel.NormalizationMethod.SOFTMAX, true, schema);
}
 

@Override
public ClusteringModel encodeModel(Schema schema){
	int[] shape = getClusterCentersShape();

	int numberOfClusters = shape[0];
	int numberOfFeatures = shape[1];

	List<? extends Number> clusterCenters = getClusterCenters();
	List<Integer> labels = getLabels();

	Multiset<Integer> labelCounts = HashMultiset.create();

	if(labels != null){
		labelCounts.addAll(labels);
	}

	List<Cluster> clusters = new ArrayList<>();

	for(int i = 0; i < numberOfClusters; i++){
		Cluster cluster = new Cluster(PMMLUtil.createRealArray(CMatrixUtil.getRow(clusterCenters, numberOfClusters, numberOfFeatures, i)))
			.setId(String.valueOf(i))
			.setSize((labelCounts.size () > 0 ? labelCounts.count(i) : null));

		clusters.add(cluster);
	}

	ComparisonMeasure comparisonMeasure = new ComparisonMeasure(ComparisonMeasure.Kind.DISTANCE, new SquaredEuclidean())
		.setCompareFunction(CompareFunction.ABS_DIFF);

	ClusteringModel clusteringModel = new ClusteringModel(MiningFunction.CLUSTERING, ClusteringModel.ModelClass.CENTER_BASED, numberOfClusters, ModelUtil.createMiningSchema(schema.getLabel()), comparisonMeasure, ClusteringModelUtil.createClusteringFields(schema.getFeatures()), clusters)
		.setOutput(ClusteringModelUtil.createOutput(FieldName.create("Cluster"), DataType.DOUBLE, clusters));

	return clusteringModel;
}
 

static
public NeuralNetwork encodeNeuralNetwork(MiningFunction miningFunction, String activation, List<? extends HasArray> coefs, List<? extends HasArray> intercepts, Schema schema){
	NeuralNetwork.ActivationFunction activationFunction = parseActivationFunction(activation);

	ClassDictUtil.checkSize(coefs, intercepts);

	Label label = schema.getLabel();
	List<? extends Feature> features = schema.getFeatures();

	NeuralInputs neuralInputs = NeuralNetworkUtil.createNeuralInputs(features, DataType.DOUBLE);

	List<? extends NeuralEntity> entities = neuralInputs.getNeuralInputs();

	List<NeuralLayer> neuralLayers = new ArrayList<>();

	for(int layer = 0; layer < coefs.size(); layer++){
		HasArray coef = coefs.get(layer);
		HasArray intercept = intercepts.get(layer);

		int[] shape = coef.getArrayShape();

		int rows = shape[0];
		int columns = shape[1];

		NeuralLayer neuralLayer = new NeuralLayer();

		List<?> coefMatrix = coef.getArrayContent();
		List<?> interceptVector = intercept.getArrayContent();

		for(int column = 0; column < columns; column++){
			List<? extends Number> weights = (List)CMatrixUtil.getColumn(coefMatrix, rows, columns, column);
			Number bias = (Number)interceptVector.get(column);

			Neuron neuron = NeuralNetworkUtil.createNeuron(entities, weights, bias)
				.setId(String.valueOf(layer + 1) + "/" + String.valueOf(column + 1));

			neuralLayer.addNeurons(neuron);
		}

		neuralLayers.add(neuralLayer);

		entities = neuralLayer.getNeurons();

		if(layer == (coefs.size() - 1)){
			neuralLayer.setActivationFunction(NeuralNetwork.ActivationFunction.IDENTITY);

			switch(miningFunction){
				case REGRESSION:
					break;
				case CLASSIFICATION:
					CategoricalLabel categoricalLabel = (CategoricalLabel)label;

					// Binary classification
					if(categoricalLabel.size() == 2){
						List<NeuralLayer> transformationNeuralLayers = NeuralNetworkUtil.createBinaryLogisticTransformation(Iterables.getOnlyElement(entities));

						neuralLayers.addAll(transformationNeuralLayers);

						neuralLayer = Iterables.getLast(transformationNeuralLayers);

						entities = neuralLayer.getNeurons();
					} else

					// Multi-class classification
					if(categoricalLabel.size() > 2){
						neuralLayer.setNormalizationMethod(NeuralNetwork.NormalizationMethod.SOFTMAX);
					} else

					{
						throw new IllegalArgumentException();
					}
					break;
				default:
					break;
			}
		}
	}

	NeuralOutputs neuralOutputs = null;

	switch(miningFunction){
		case REGRESSION:
			neuralOutputs = NeuralNetworkUtil.createRegressionNeuralOutputs(entities, (ContinuousLabel)label);
			break;
		case CLASSIFICATION:
			neuralOutputs = NeuralNetworkUtil.createClassificationNeuralOutputs(entities, (CategoricalLabel)label);
			break;
		default:
			break;
	}

	NeuralNetwork neuralNetwork = new NeuralNetwork(miningFunction, activationFunction, ModelUtil.createMiningSchema(label), neuralInputs, neuralLayers)
		.setNeuralOutputs(neuralOutputs);

	return neuralNetwork;
}
 

private Model encodeMultinomialModel(Schema schema){
	String sklearnVersion = getSkLearnVersion();
	int[] shape = getCoefShape();

	int numberOfClasses = shape[0];
	int numberOfFeatures = shape[1];

	List<? extends Number> coef = getCoef();
	List<? extends Number> intercept = getIntercept();

	CategoricalLabel categoricalLabel = (CategoricalLabel)schema.getLabel();

	List<? extends Feature> features = schema.getFeatures();

	// See https://github.com/scikit-learn/scikit-learn/issues/6848
	boolean corrected = (sklearnVersion != null && SkLearnUtil.compareVersion(sklearnVersion, "0.21") >= 0);

	if(!corrected){
		return super.encodeModel(schema);
	} // End if

	if(numberOfClasses >= 3){
		SchemaUtil.checkSize(numberOfClasses, categoricalLabel);

		Schema segmentSchema = (schema.toAnonymousRegressorSchema(DataType.DOUBLE)).toEmptySchema();

		List<RegressionModel> regressionModels = new ArrayList<>();

		for(int i = 0, rows = categoricalLabel.size(); i < rows; i++){
			RegressionModel regressionModel = RegressionModelUtil.createRegression(features, CMatrixUtil.getRow(coef, numberOfClasses, numberOfFeatures, i), intercept.get(i), RegressionModel.NormalizationMethod.NONE, segmentSchema)
				.setOutput(ModelUtil.createPredictedOutput(FieldName.create("decisionFunction(" + categoricalLabel.getValue(i) + ")"), OpType.CONTINUOUS, DataType.DOUBLE));

			regressionModels.add(regressionModel);
		}

		return MiningModelUtil.createClassification(regressionModels, RegressionModel.NormalizationMethod.SOFTMAX, true, schema);
	} else

	{
		throw new IllegalArgumentException();
	}
}
 

private MiningModel encodeMultinomialClassification(List<TreeModel> treeModels, Double initF, Schema schema){
	CategoricalLabel categoricalLabel = (CategoricalLabel)schema.getLabel();

	Schema segmentSchema = schema.toAnonymousRegressorSchema(DataType.DOUBLE);

	List<Model> miningModels = new ArrayList<>();

	for(int i = 0, columns = categoricalLabel.size(), rows = (treeModels.size() / columns); i < columns; i++){
		MiningModel miningModel = createMiningModel(CMatrixUtil.getColumn(treeModels, rows, columns, i), initF, segmentSchema)
			.setOutput(ModelUtil.createPredictedOutput(FieldName.create("gbmValue(" + categoricalLabel.getValue(i) + ")"), OpType.CONTINUOUS, DataType.DOUBLE));

		miningModels.add(miningModel);
	}

	return MiningModelUtil.createClassification(miningModels, RegressionModel.NormalizationMethod.SOFTMAX, true, schema);
}
 

@Override
public NaiveBayesModel encodeModel(Schema schema){
	int[] shape = getThetaShape();

	int numberOfClasses = shape[0];
	int numberOfFeatures = shape[1];

	List<? extends Number> theta = getTheta();
	List<? extends Number> sigma = getSigma();

	CategoricalLabel categoricalLabel = (CategoricalLabel)schema.getLabel();

	BayesInputs bayesInputs = new BayesInputs();

	for(int i = 0; i < numberOfFeatures; i++){
		Feature feature = schema.getFeature(i);

		List<? extends Number> means = CMatrixUtil.getColumn(theta, numberOfClasses, numberOfFeatures, i);
		List<? extends Number> variances = CMatrixUtil.getColumn(sigma, numberOfClasses, numberOfFeatures, i);

		ContinuousFeature continuousFeature = feature.toContinuousFeature();

		BayesInput bayesInput = new BayesInput(continuousFeature.getName(), encodeTargetValueStats(categoricalLabel.getValues(), means, variances), null);

		bayesInputs.addBayesInputs(bayesInput);
	}

	List<Integer> classCount = getClassCount();

	BayesOutput bayesOutput = new BayesOutput(categoricalLabel.getName(), null)
		.setTargetValueCounts(encodeTargetValueCounts(categoricalLabel.getValues(), classCount));

	NaiveBayesModel naiveBayesModel = new NaiveBayesModel(0d, MiningFunction.CLASSIFICATION, ModelUtil.createMiningSchema(categoricalLabel), bayesInputs, bayesOutput)
		.setOutput(ModelUtil.createProbabilityOutput(DataType.DOUBLE, categoricalLabel));

	return naiveBayesModel;
}
 

@Override
public Model encodeModel(Schema schema){
	int[] shape = getCoefShape();

	int numberOfClasses = shape[0];
	int numberOfFeatures = shape[1];

	boolean hasProbabilityDistribution = hasProbabilityDistribution();

	List<? extends Number> coef = getCoef();
	List<? extends Number> intercept = getIntercept();

	CategoricalLabel categoricalLabel = (CategoricalLabel)schema.getLabel();

	List<? extends Feature> features = schema.getFeatures();

	if(numberOfClasses == 1){
		SchemaUtil.checkSize(2, categoricalLabel);

		return RegressionModelUtil.createBinaryLogisticClassification(features, CMatrixUtil.getRow(coef, numberOfClasses, numberOfFeatures, 0), intercept.get(0), RegressionModel.NormalizationMethod.LOGIT, hasProbabilityDistribution, schema);
	} else

	if(numberOfClasses >= 3){
		SchemaUtil.checkSize(numberOfClasses, categoricalLabel);

		Schema segmentSchema = (schema.toAnonymousRegressorSchema(DataType.DOUBLE)).toEmptySchema();

		List<RegressionModel> regressionModels = new ArrayList<>();

		for(int i = 0, rows = categoricalLabel.size(); i < rows; i++){
			RegressionModel regressionModel = RegressionModelUtil.createRegression(features, CMatrixUtil.getRow(coef, numberOfClasses, numberOfFeatures, i), intercept.get(i), RegressionModel.NormalizationMethod.LOGIT, segmentSchema)
				.setOutput(ModelUtil.createPredictedOutput(FieldName.create("decisionFunction(" + categoricalLabel.getValue(i) + ")"), OpType.CONTINUOUS, DataType.DOUBLE));

			regressionModels.add(regressionModel);
		}

		return MiningModelUtil.createClassification(regressionModels, RegressionModel.NormalizationMethod.SIMPLEMAX, hasProbabilityDistribution, schema);
	} else

	{
		throw new IllegalArgumentException();
	}
}
 

@Override
public List<Feature> encodeFeatures(List<Feature> features, SkLearnEncoder encoder){
	int[] shape = getComponentsShape();

	int numberOfComponents = shape[0];
	int numberOfFeatures = shape[1];

	List<? extends Number> components = getComponents();
	List<? extends Number> mean = getMean();

	ClassDictUtil.checkSize(numberOfFeatures, features, mean);

	Boolean whiten = getWhiten();

	List<? extends Number> explainedVariance = (whiten ? getExplainedVariance() : null);

	ClassDictUtil.checkSize(numberOfComponents, explainedVariance);

	String id = "pca@" + String.valueOf(PCA.SEQUENCE.getAndIncrement());

	List<Feature> result = new ArrayList<>();

	for(int i = 0; i < numberOfComponents; i++){
		List<? extends Number> component = CMatrixUtil.getRow(components, numberOfComponents, numberOfFeatures, i);

		Apply apply = PMMLUtil.createApply(PMMLFunctions.SUM);

		for(int j = 0; j < numberOfFeatures; j++){
			Feature feature = features.get(j);

			Number meanValue = mean.get(j);
			Number componentValue = component.get(j);

			if(ValueUtil.isZero(meanValue) && ValueUtil.isOne(componentValue)){
				apply.addExpressions(feature.ref());

				continue;
			}

			ContinuousFeature continuousFeature = feature.toContinuousFeature();

			// "($name[i] - mean[i]) * component[i]"
			Expression expression = continuousFeature.ref();

			if(!ValueUtil.isZero(meanValue)){
				expression = PMMLUtil.createApply(PMMLFunctions.SUBTRACT, expression, PMMLUtil.createConstant(meanValue));
			} // End if

			if(!ValueUtil.isOne(componentValue)){
				expression = PMMLUtil.createApply(PMMLFunctions.MULTIPLY, expression, PMMLUtil.createConstant(componentValue));
			}

			apply.addExpressions(expression);
		}

		if(whiten){
			Number explainedVarianceValue = explainedVariance.get(i);

			if(!ValueUtil.isOne(explainedVarianceValue)){
				apply = PMMLUtil.createApply(PMMLFunctions.DIVIDE, apply, PMMLUtil.createConstant(Math.sqrt(ValueUtil.asDouble(explainedVarianceValue))));
			}
		}

		DerivedField derivedField = encoder.createDerivedField(FieldName.create(id + "[" + String.valueOf(i) + "]"), apply);

		result.add(new ContinuousFeature(encoder, derivedField));
	}

	return result;
}
 

@Override
public List<Feature> encodeFeatures(List<Feature> features, SkLearnEncoder encoder){
	int[] shape = getComponentsShape();

	int numberOfComponents = shape[0];
	int numberOfFeatures = shape[1];

	List<? extends Number> components = getComponents();

	ClassDictUtil.checkSize(numberOfFeatures, features);

	String id = "svd@" + String.valueOf(TruncatedSVD.SEQUENCE.getAndIncrement());

	List<Feature> result = new ArrayList<>();

	for(int i = 0; i < numberOfComponents; i++){
		List<? extends Number> component = CMatrixUtil.getRow(components, numberOfComponents, numberOfFeatures, i);

		Apply apply = PMMLUtil.createApply(PMMLFunctions.SUM);

		for(int j = 0; j < numberOfFeatures; j++){
			Feature feature = features.get(j);

			Number componentValue = component.get(j);

			if(ValueUtil.isOne(componentValue)){
				apply.addExpressions(feature.ref());

				continue;
			}

			ContinuousFeature continuousFeature = feature.toContinuousFeature();

			// "$name[i] * component[i]"
			Expression expression = PMMLUtil.createApply(PMMLFunctions.MULTIPLY, continuousFeature.ref(), PMMLUtil.createConstant(componentValue));

			apply.addExpressions(expression);
		}

		DerivedField derivedField = encoder.createDerivedField(FieldName.create(id + "[" + String.valueOf(i) + "]"), apply);

		result.add(new ContinuousFeature(encoder, derivedField));
	}

	return result;
}