Example usage for org.deeplearning4j.nn.multilayer MultiLayerNetwork score

Introduction

In this page you can find the example usage for org.deeplearning4j.nn.multilayer MultiLayerNetwork score.

Prototype

double score

To view the source code for org.deeplearning4j.nn.multilayer MultiLayerNetwork score.

Click Source Link

Usage

From source file:org.knime.ext.dl4j.base.nodes.learn.AbstractDLLearnerNodeModel.java

License:Open Source License

/**
 * Logs score of specified model at specified epoch in the view and adds the information to the history.
 *
 * @param m the model to get score from/*  w w  w  .  j a  v  a 2s  .co m*/
 * @param epoch the epoch number to print into log message
 */
protected void logEpochScore(final MultiLayerNetwork m, final int epoch) {
    HistoryEntry entry = new HistoryEntry(m.score(), epoch);
    m_history.add(entry);
    notifyViews(entry);
}

---

From source file:org.knime.ext.dl4j.base.nodes.predict.feedforward.FeedforwardPredictorNodeModel.java

License:Open Source License

@Override
protected PortObject[] execute(final PortObject[] inObjects, final ExecutionContext exec) throws Exception {
    final DLModelPortObject port = (DLModelPortObject) inObjects[0];
    final DLModelPortObjectSpec portSpec = (DLModelPortObjectSpec) port.getSpec();
    final BufferedDataTable table = (BufferedDataTable) inObjects[1];

    //select feature columns from table used for prediction
    final String[] predictCols = DLModelPortObjectUtils.getFirsts(portSpec.getLearnedColumns(), String.class);
    final BufferedDataTable filteredTable = exec
            .createBufferedDataTable(new FilterColumnTable(table, predictCols), exec);

    //create iterator and prediction
    TableUtils.checkForEmptyTable(filteredTable);
    final DataSetIterator input = new ClassificationBufferedDataTableDataSetIterator(filteredTable, 1);
    final MultiLayerNetwork mln = port.getMultilayerLayerNetwork();

    //set flag if last layer activation is softmax
    final boolean outputActivationIsSoftmax = isOutActivation(port.getLayers(), DL4JActivationFunction.softmax);

    final boolean appendPrediction = m_predictorParameter.getAppendPrediction().getBooleanValue();
    final boolean appendScore = m_predictorParameter.getAppendScore().getBooleanValue();
    final List<String> labels = portSpec.getLabels();

    //write output to table
    final BufferedDataContainer container = exec.createDataContainer(m_outputSpec);
    final CloseableRowIterator tabelIter = table.iterator();

    int i = 0;//from  w  w w  .ja va 2 s  . com
    while (tabelIter.hasNext()) {
        exec.setProgress((double) (i + 1) / (double) (table.size()));
        exec.checkCanceled();

        final DataRow row = tabelIter.next();
        final List<DataCell> cells = TableUtils.toListOfCells(row);

        final DataSet next = input.next();
        final INDArray prediction = predict(mln, next.getFeatureMatrix());

        final ListCell outputVector = CollectionCellFactory
                .createListCell(NDArrayUtils.toListOfDoubleCells(prediction));
        cells.add(outputVector);
        if (appendScore) {
            final double score = mln.score(new DataSet(next.getFeatureMatrix(), prediction), false);
            cells.add(new DoubleCell(score));
        }
        if (appendPrediction && outputActivationIsSoftmax && containsLabels()) {
            final String winningLabel = NDArrayUtils.softmaxActivationToLabel(labels, prediction);
            cells.add(new StringCell(winningLabel));
        } else if (appendPrediction && containsLabels()) {
            cells.add(new MissingCell("Output Layer activation is not softmax"));
        } else if (appendPrediction && !containsLabels()) {
            cells.add(new MissingCell("Model contains no labels"));
        }

        container.addRowToTable(new DefaultRow(row.getKey(), cells));
        i++;
    }
    if (appendPrediction && !outputActivationIsSoftmax) {
        logger.warn("Output Layer activation is not softmax. Label prediction column will be empty.");
    }
    if (appendPrediction && outputActivationIsSoftmax && !containsLabels()) {
        logger.warn(
                "Model contains no labels. May be trained unsupervised. Label prediction column will be empty.");
    }

    container.close();
    final BufferedDataTable outputTable = container.getTable();

    return new PortObject[] { outputTable };
}

---