Examples with TreeEnsembleModel org.knime.base.node.mine.treeensemble.model.TreeEnsembleModel used on opensource projects

Example 6 with TreeEnsembleModel

use of org.knime.base.node.mine.treeensemble.model.TreeEnsembleModel in project knime-core by knime.

the class TreeEnsembleShrinkerNodeModel method execute.

@Override
protected PortObject[] execute(final PortObject[] inObjects, final ExecutionContext exec) throws Exception {
    TreeEnsembleModel treeEnsemble = ((TreeEnsembleModelPortObject) inObjects[0]).getEnsembleModel();
    TreeEnsembleModelPortObject resultEnsemble;
    int resultSize = m_config.getResultSize(treeEnsemble.getNrModels());
    boolean shrink = true;
    if (!m_config.isResultSizeAutomatic()) {
        // Check if result size is valid
        if (resultSize < 1) {
            // Result size is to small, use 1
            setWarningMessage("The configured result size is smaller than 1, defaulting to 1");
            resultSize = 1;
        } else if (resultSize > treeEnsemble.getNrModels()) {
            // Result size is to big, just keep current ensemble
            setWarningMessage("The configured result size is bigger than the size of the input ensemble, defaulting to the input ensembles size");
            shrink = false;
        } else if (resultSize == treeEnsemble.getNrModels()) {
            // Result size is ensemble size -> we don't need to shrink
            shrink = false;
        }
    }
    // If our result size is not smaller than the current ensemble we don't have to do the following and therefore can save time
    if (shrink) {
        BufferedDataTable inData = (BufferedDataTable) inObjects[1];
        // Create shrinker
        TreeEnsembleShrinker shrinker = new TreeEnsembleShrinker(treeEnsemble, inData, m_config.getTargetColumn(), exec);
        // Shrink ensemble
        if (m_config.isResultSizeAutomatic()) {
            shrinker.autoShrink();
        } else {
            shrinker.shrinkTo(resultSize);
        }
        // Get shrunk ensemble
        TreeEnsembleModel newEnsemble = shrinker.getModel();
        // Push flow variable with archived accuracy
        pushFlowVariableDouble("Tree Ensemble Shrinker Prediction Accuracy", shrinker.getAccuracy());
        // Create port object for tree ensemble
        resultEnsemble = new TreeEnsembleModelPortObject(((TreeEnsembleModelPortObject) inObjects[0]).getSpec(), newEnsemble);
    } else {
        // We did not need to shrink just use input tree ensemble port object
        resultEnsemble = (TreeEnsembleModelPortObject) inObjects[0];
    }
    // Convert tree ensemble port object to PMML
    PMMLPortObject pmmlEnsemble = convertToPmmlEnsemble(resultEnsemble, exec);
    return new PortObject[] { pmmlEnsemble };
}

Example 7 with TreeEnsembleModel

use of org.knime.base.node.mine.treeensemble.model.TreeEnsembleModel in project knime-core by knime.

the class TreeEnsembleLearner method learnEnsemble.

public TreeEnsembleModel learnEnsemble(final ExecutionMonitor exec) throws CanceledExecutionException, ExecutionException {
    final int nrModels = m_config.getNrModels();
    final RandomData rd = m_config.createRandomData();
    final ThreadPool tp = KNIMEConstants.GLOBAL_THREAD_POOL;
    final AtomicReference<Throwable> learnThrowableRef = new AtomicReference<Throwable>();
    @SuppressWarnings("unchecked") final Future<TreeLearnerResult>[] modelFutures = new Future[nrModels];
    final int procCount = 3 * Runtime.getRuntime().availableProcessors() / 2;
    final Semaphore semaphore = new Semaphore(procCount);
    Callable<TreeLearnerResult[]> learnCallable = new Callable<TreeLearnerResult[]>() {

        @Override
        public TreeLearnerResult[] call() throws Exception {
            final TreeLearnerResult[] results = new TreeLearnerResult[nrModels];
            for (int i = 0; i < nrModels; i++) {
                semaphore.acquire();
                finishedTree(i - procCount, exec);
                checkThrowable(learnThrowableRef);
                RandomData rdSingle = TreeEnsembleLearnerConfiguration.createRandomData(rd.nextLong(Long.MIN_VALUE, Long.MAX_VALUE));
                ExecutionMonitor subExec = exec.createSubProgress(0.0);
                modelFutures[i] = tp.enqueue(new TreeLearnerCallable(subExec, rdSingle, learnThrowableRef, semaphore));
            }
            for (int i = 0; i < procCount; i++) {
                semaphore.acquire();
                finishedTree(nrModels - 1 + i - procCount, exec);
            }
            for (int i = 0; i < nrModels; i++) {
                try {
                    results[i] = modelFutures[i].get();
                } catch (Exception e) {
                    learnThrowableRef.compareAndSet(null, e);
                }
            }
            return results;
        }

        private void finishedTree(final int treeIndex, final ExecutionMonitor progMon) {
            if (treeIndex > 0) {
                progMon.setProgress(treeIndex / (double) nrModels, "Tree " + treeIndex + "/" + nrModels);
            }
        }
    };
    TreeLearnerResult[] modelResults = tp.runInvisible(learnCallable);
    checkThrowable(learnThrowableRef);
    AbstractTreeModel[] models = new AbstractTreeModel[nrModels];
    m_rowSamples = new RowSample[nrModels];
    m_columnSampleStrategies = new ColumnSampleStrategy[nrModels];
    for (int i = 0; i < nrModels; i++) {
        models[i] = modelResults[i].m_treeModel;
        m_rowSamples[i] = modelResults[i].m_rowSample;
        m_columnSampleStrategies[i] = modelResults[i].m_rootColumnSampleStrategy;
    }
    m_ensembleModel = new TreeEnsembleModel(m_config, m_data.getMetaData(), models, m_data.getTreeType());
    return m_ensembleModel;
}

Example 8 with TreeEnsembleModel

use of org.knime.base.node.mine.treeensemble.model.TreeEnsembleModel in project knime-core by knime.

the class TreeEnsembleRegressionLearnerNodeModel method execute.

/**
 * {@inheritDoc}
 */
@Override
protected PortObject[] execute(final PortObject[] inObjects, final ExecutionContext exec) throws Exception {
    BufferedDataTable t = (BufferedDataTable) inObjects[0];
    DataTableSpec spec = t.getDataTableSpec();
    final FilterLearnColumnRearranger learnRearranger = m_configuration.filterLearnColumns(spec);
    String warn = learnRearranger.getWarning();
    BufferedDataTable learnTable = exec.createColumnRearrangeTable(t, learnRearranger, exec.createSubProgress(0.0));
    DataTableSpec learnSpec = learnTable.getDataTableSpec();
    TreeEnsembleModelPortObjectSpec ensembleSpec = m_configuration.createPortObjectSpec(learnSpec);
    ExecutionMonitor readInExec = exec.createSubProgress(0.1);
    ExecutionMonitor learnExec = exec.createSubProgress(0.8);
    ExecutionMonitor outOfBagExec = exec.createSubProgress(0.1);
    TreeDataCreator dataCreator = new TreeDataCreator(m_configuration, learnSpec, learnTable.getRowCount());
    exec.setProgress("Reading data into memory");
    TreeData data = dataCreator.readData(learnTable, m_configuration, readInExec);
    m_hiliteRowSample = dataCreator.getDataRowsForHilite();
    m_viewMessage = dataCreator.getViewMessage();
    String dataCreationWarning = dataCreator.getAndClearWarningMessage();
    if (dataCreationWarning != null) {
        if (warn == null) {
            warn = dataCreationWarning;
        } else {
            warn = warn + "\n" + dataCreationWarning;
        }
    }
    readInExec.setProgress(1.0);
    exec.setMessage("Learning trees");
    TreeEnsembleLearner learner = new TreeEnsembleLearner(m_configuration, data);
    TreeEnsembleModel model;
    try {
        model = learner.learnEnsemble(learnExec);
    } catch (ExecutionException e) {
        Throwable cause = e.getCause();
        if (cause instanceof Exception) {
            throw (Exception) cause;
        }
        throw e;
    }
    TreeEnsembleModelPortObject modelPortObject = new TreeEnsembleModelPortObject(ensembleSpec, model);
    learnExec.setProgress(1.0);
    exec.setMessage("Out of bag prediction");
    TreeEnsemblePredictor outOfBagPredictor = createOutOfBagPredictor(ensembleSpec, modelPortObject, spec);
    outOfBagPredictor.setOutofBagFilter(learner.getRowSamples(), data.getTargetColumn());
    ColumnRearranger outOfBagRearranger = outOfBagPredictor.getPredictionRearranger();
    BufferedDataTable outOfBagTable = exec.createColumnRearrangeTable(t, outOfBagRearranger, outOfBagExec);
    BufferedDataTable colStatsTable = learner.createColumnStatisticTable(exec.createSubExecutionContext(0.0));
    m_ensembleModelPortObject = modelPortObject;
    if (warn != null) {
        setWarningMessage(warn);
    }
    return new PortObject[] { outOfBagTable, colStatsTable, modelPortObject };
}

Example 9 with TreeEnsembleModel

use of org.knime.base.node.mine.treeensemble.model.TreeEnsembleModel in project knime-core by knime.

the class TreeEnsembleRegressionPredictorCellFactory method getCells.

/**
 * {@inheritDoc}
 */
@Override
public DataCell[] getCells(final DataRow row) {
    TreeEnsembleModelPortObject modelObject = m_predictor.getModelObject();
    TreeEnsemblePredictorConfiguration cfg = m_predictor.getConfiguration();
    final TreeEnsembleModel ensembleModel = modelObject.getEnsembleModel();
    int size = 1;
    final boolean appendConfidence = cfg.isAppendPredictionConfidence();
    final boolean appendModelCount = cfg.isAppendModelCount();
    if (appendConfidence) {
        size += 1;
    }
    if (appendModelCount) {
        size += 1;
    }
    final boolean hasOutOfBagFilter = m_predictor.hasOutOfBagFilter();
    DataCell[] result = new DataCell[size];
    DataRow filterRow = new FilterColumnRow(row, m_learnColumnInRealDataIndices);
    PredictorRecord record = ensembleModel.createPredictorRecord(filterRow, m_learnSpec);
    if (record == null) {
        // missing value
        Arrays.fill(result, DataType.getMissingCell());
        return result;
    }
    Mean mean = new Mean();
    Variance variance = new Variance();
    final int nrModels = ensembleModel.getNrModels();
    for (int i = 0; i < nrModels; i++) {
        if (hasOutOfBagFilter && m_predictor.isRowPartOfTrainingData(row.getKey(), i)) {
        // ignore, row was used to train the model
        } else {
            TreeModelRegression m = ensembleModel.getTreeModelRegression(i);
            TreeNodeRegression match = m.findMatchingNode(record);
            double nodeMean = match.getMean();
            mean.increment(nodeMean);
            variance.increment(nodeMean);
        }
    }
    int nrValidModels = (int) mean.getN();
    int index = 0;
    result[index++] = nrValidModels == 0 ? DataType.getMissingCell() : new DoubleCell(mean.getResult());
    if (appendConfidence) {
        result[index++] = nrValidModels == 0 ? DataType.getMissingCell() : new DoubleCell(variance.getResult());
    }
    if (appendModelCount) {
        result[index++] = new IntCell(nrValidModels);
    }
    return result;
}

Example 10 with TreeEnsembleModel

use of org.knime.base.node.mine.treeensemble.model.TreeEnsembleModel in project knime-core by knime.

the class TreeEnsembleLearnerNodeView method newModel.

private void newModel(final int index) {
    assert SwingUtilities.isEventDispatchThread();
    final MODEL nodeModel = getNodeModel();
    TreeEnsembleModel model = nodeModel.getEnsembleModel();
    DataTable hiliteRowSample = nodeModel.getHiliteRowSample();
    UpdateTreeWorker updateWorker = new UpdateTreeWorker(hiliteRowSample, model, index);
    UpdateTreeWorker old = m_updateWorkerRef.getAndSet(updateWorker);
    if (old != null) {
        old.cancel(true);
    }
    updateWorker.execute();
}