Examples with MultiLayerPerceptron org.knime.base.data.neural.MultiLayerPerceptron used on opensource projects

Example 6 with MultiLayerPerceptron

use of org.knime.base.data.neural.MultiLayerPerceptron in project knime-core by knime.

the class PMMLNeuralNetworkTranslator method addOutputLayer.

/**
 * Writes the PMML output layer of the MLP.
 *
 * @param nnModel
 *            the neural network model.
 * @param mlp
 *            the underlying {@link MultiLayerPerceptron}.
 * @param spec
 *            the port object spec
 */
protected void addOutputLayer(final NeuralNetwork nnModel, final MultiLayerPerceptron mlp, final PMMLPortObjectSpec spec) {
    int lastlayer = mlp.getNrLayers() - 1;
    String targetCol = spec.getTargetFields().iterator().next();
    Layer outputlayer = mlp.getLayer(lastlayer);
    Perceptron[] outputperceptrons = outputlayer.getPerceptrons();
    HashMap<DataCell, Integer> outputmap = mlp.getClassMapping();
    NeuralOutputs neuralOuts = nnModel.addNewNeuralOutputs();
    neuralOuts.setNumberOfOutputs(BigInteger.valueOf(outputperceptrons.length));
    for (int i = 0; i < outputperceptrons.length; i++) {
        NeuralOutput neuralOutput = neuralOuts.addNewNeuralOutput();
        neuralOutput.setOutputNeuron(lastlayer + "," + i);
        // search corresponding output value
        String colname = "";
        for (Entry<DataCell, Integer> e : outputmap.entrySet()) {
            if (e.getValue().equals(i)) {
                colname = ((StringValue) e.getKey()).getStringValue();
            }
        }
        DerivedField df = neuralOutput.addNewDerivedField();
        df.setOptype(OPTYPE.CATEGORICAL);
        df.setDataType(DATATYPE.STRING);
        if (mlp.getMode() == MultiLayerPerceptron.CLASSIFICATION_MODE) {
            df.setOptype(OPTYPE.CATEGORICAL);
            df.setDataType(DATATYPE.STRING);
        } else if (mlp.getMode() == MultiLayerPerceptron.REGRESSION_MODE) {
            df.setOptype(OPTYPE.CONTINUOUS);
            df.setDataType(DATATYPE.DOUBLE);
        }
        if (mlp.getMode() == MultiLayerPerceptron.CLASSIFICATION_MODE) {
            NormDiscrete normDiscrete = df.addNewNormDiscrete();
            normDiscrete.setField(targetCol);
            normDiscrete.setValue(colname);
        } else if (mlp.getMode() == MultiLayerPerceptron.REGRESSION_MODE) {
            FieldRef fieldRef = df.addNewFieldRef();
            fieldRef.setField(targetCol);
        }
    }
}

Example 7 with MultiLayerPerceptron

use of org.knime.base.data.neural.MultiLayerPerceptron in project knime-core by knime.

the class PMMLNeuralNetworkTranslator method initializeFrom.

/**
 * {@inheritDoc}
 */
@Override
public void initializeFrom(final PMMLDocument pmmlDoc) {
    m_nameMapper = new DerivedFieldMapper(pmmlDoc);
    NeuralNetwork[] models = pmmlDoc.getPMML().getNeuralNetworkArray();
    if (models.length == 0) {
        throw new IllegalArgumentException("No neural network model" + " provided.");
    } else if (models.length > 1) {
        LOGGER.warn("Multiple neural network models found. " + "Only the first model is considered.");
    }
    NeuralNetwork nnModel = models[0];
    // ------------------------------
    // initiate Neural Input
    initInputLayer(nnModel);
    // -------------------------------
    // initiate Hidden Layer
    initiateHiddenLayers(nnModel);
    // -------------------------------
    // initiate Final Layer
    initiateFinalLayer(nnModel);
    // --------------------------------
    // initiate Neural Outputs
    initiateNeuralOutputs(nnModel);
    // --------------------------------
    // initiate Neural Network properties
    ACTIVATIONFUNCTION.Enum actFunc = nnModel.getActivationFunction();
    NNNORMALIZATIONMETHOD.Enum normMethod = nnModel.getNormalizationMethod();
    if (ACTIVATIONFUNCTION.LOGISTIC != actFunc) {
        LOGGER.error("Only logistic activation function is " + "supported in KNIME MLP.");
    }
    if (NNNORMALIZATIONMETHOD.NONE != normMethod) {
        LOGGER.error("No normalization method is " + "supported in KNIME MLP.");
    }
    MININGFUNCTION.Enum functionName = nnModel.getFunctionName();
    if (MININGFUNCTION.CLASSIFICATION == functionName) {
        m_mlpMethod = MultiLayerPerceptron.CLASSIFICATION_MODE;
    } else if (MININGFUNCTION.REGRESSION == functionName) {
        m_mlpMethod = MultiLayerPerceptron.REGRESSION_MODE;
    }
    if (m_allLayers.size() < 3) {
        throw new IllegalArgumentException("Only neural networks with 3 Layers supported in KNIME MLP.");
    }
    Layer[] allLayers = new Layer[m_allLayers.size()];
    allLayers = m_allLayers.toArray(allLayers);
    m_mlp = new MultiLayerPerceptron(allLayers);
    Architecture myarch = new Architecture(allLayers[0].getPerceptrons().length, allLayers.length - 2, allLayers[1].getPerceptrons().length, allLayers[allLayers.length - 1].getPerceptrons().length);
    m_mlp.setArchitecture(myarch);
    m_mlp.setClassMapping(m_classmap);
    m_mlp.setInputMapping(m_inputmap);
    m_mlp.setMode(m_mlpMethod);
}

Example 8 with MultiLayerPerceptron

use of org.knime.base.data.neural.MultiLayerPerceptron in project knime-core by knime.

the class PMMLNeuralNetworkTranslator method addLayer.

/**
 * Writes a layer of the MLP.
 *
 * @param nnModel
 *            the NeuralNetwork model.
 * @param mlp
 *            the underlying {@link MultiLayerPerceptron}.
 * @param layer
 *            the number of the current layer.
 */
protected void addLayer(final NeuralNetwork nnModel, final MultiLayerPerceptron mlp, final int layer) {
    Layer curLayer = mlp.getLayer(layer);
    Perceptron[] perceptrons = curLayer.getPerceptrons();
    AttributesImpl atts = new AttributesImpl();
    atts.addAttribute(null, null, "numberOfNeurons", CDATA, "" + perceptrons.length);
    NeuralLayer neuralLayer = nnModel.addNewNeuralLayer();
    for (int i = 0; i < perceptrons.length; i++) {
        Neuron neuron = neuralLayer.addNewNeuron();
        neuron.setId(layer + "," + i);
        neuron.setBias(-1 * perceptrons[i].getThreshold());
        double[] weights = perceptrons[i].getWeights();
        int predLayerLength = weights.length;
        for (int j = 0; j < predLayerLength; j++) {
            Con con = neuron.addNewCon();
            con.setFrom((layer - 1) + "," + j);
            con.setWeight(weights[j]);
        }
    }
}

Example 9 with MultiLayerPerceptron

use of org.knime.base.data.neural.MultiLayerPerceptron in project knime-core by knime.

the class RPropNodeModel method execute.

/**
 * The execution consists of three steps:
 * <ol>
 * <li>A neural network is build with the inputs and outputs according to
 * the input datatable, number of hidden layers as specified.</li>
 * <li>Input DataTables are converted into double-arrays so they can be
 * attached to the neural net.</li>
 * <li>The neural net is trained.</li>
 * </ol>
 *
 * {@inheritDoc}
 */
@Override
protected PortObject[] execute(final PortObject[] inData, final ExecutionContext exec) throws Exception {
    // If class column is not set, it is the last column.
    DataTableSpec posSpec = (DataTableSpec) inData[INDATA].getSpec();
    if (m_classcol.getStringValue() == null) {
        m_classcol.setStringValue(posSpec.getColumnSpec(posSpec.getNumColumns() - 1).getName());
    }
    List<String> learningCols = new LinkedList<String>();
    List<String> targetCols = new LinkedList<String>();
    // Determine the number of inputs and the number of outputs. Make also
    // sure that the inputs are double values.
    int nrInputs = 0;
    int nrOutputs = 0;
    HashMap<String, Integer> inputmap = new HashMap<String, Integer>();
    HashMap<DataCell, Integer> classMap = new HashMap<DataCell, Integer>();
    for (DataColumnSpec colspec : posSpec) {
        // check for class column
        if (colspec.getName().toString().compareTo(m_classcol.getStringValue()) == 0) {
            targetCols.add(colspec.getName());
            if (colspec.getType().isCompatible(DoubleValue.class)) {
                // check if the values are in range [0,1]
                DataColumnDomain domain = colspec.getDomain();
                if (domain.hasBounds()) {
                    double lower = ((DoubleValue) domain.getLowerBound()).getDoubleValue();
                    double upper = ((DoubleValue) domain.getUpperBound()).getDoubleValue();
                    if (lower < 0 || upper > 1) {
                        throw new InvalidSettingsException("Domain range for regression in column " + colspec.getName() + " not in range [0,1]");
                    }
                }
                nrOutputs = 1;
                classMap = new HashMap<DataCell, Integer>();
                classMap.put(new StringCell(colspec.getName()), 0);
                m_regression = true;
            } else {
                m_regression = false;
                DataColumnDomain domain = colspec.getDomain();
                if (domain.hasValues()) {
                    Set<DataCell> allvalues = domain.getValues();
                    int outputneuron = 0;
                    classMap = new HashMap<DataCell, Integer>();
                    for (DataCell value : allvalues) {
                        classMap.put(value, outputneuron);
                        outputneuron++;
                    }
                    nrOutputs = allvalues.size();
                } else {
                    throw new Exception("Could not find domain values in" + "nominal column " + colspec.getName().toString());
                }
            }
        } else {
            if (!colspec.getType().isCompatible(DoubleValue.class)) {
                throw new Exception("Only double columns for input");
            }
            inputmap.put(colspec.getName(), nrInputs);
            learningCols.add(colspec.getName());
            nrInputs++;
        }
    }
    assert targetCols.size() == 1 : "Only one class column allowed.";
    m_architecture.setNrInputNeurons(nrInputs);
    m_architecture.setNrHiddenLayers(m_nrHiddenLayers.getIntValue());
    m_architecture.setNrHiddenNeurons(m_nrHiddenNeuronsperLayer.getIntValue());
    m_architecture.setNrOutputNeurons(nrOutputs);
    Random random = new Random();
    if (m_useRandomSeed.getBooleanValue()) {
        random.setSeed(m_randomSeed.getIntValue());
    }
    m_mlp = new MultiLayerPerceptron(m_architecture, random);
    if (m_regression) {
        m_mlp.setMode(MultiLayerPerceptron.REGRESSION_MODE);
    } else {
        m_mlp.setMode(MultiLayerPerceptron.CLASSIFICATION_MODE);
    }
    // Convert inputs to double arrays. Values from the class column are
    // encoded as bitvectors.
    int classColNr = posSpec.findColumnIndex(m_classcol.getStringValue());
    List<Double[]> samples = new ArrayList<Double[]>();
    List<Double[]> outputs = new ArrayList<Double[]>();
    Double[] sample = new Double[nrInputs];
    Double[] output = new Double[nrOutputs];
    final RowIterator rowIt = ((BufferedDataTable) inData[INDATA]).iterator();
    int rowcounter = 0;
    while (rowIt.hasNext()) {
        boolean add = true;
        output = new Double[nrOutputs];
        sample = new Double[nrInputs];
        DataRow row = rowIt.next();
        int nrCells = row.getNumCells();
        int index = 0;
        for (int i = 0; i < nrCells; i++) {
            if (i != classColNr) {
                if (!row.getCell(i).isMissing()) {
                    DoubleValue dc = (DoubleValue) row.getCell(i);
                    sample[index] = dc.getDoubleValue();
                    index++;
                } else {
                    if (m_ignoreMV.getBooleanValue()) {
                        add = false;
                        break;
                    } else {
                        throw new Exception("Missing values in input" + " datatable");
                    }
                }
            } else {
                if (row.getCell(i).isMissing()) {
                    add = false;
                    if (!m_ignoreMV.getBooleanValue()) {
                        throw new Exception("Missing value in class" + " column");
                    }
                    break;
                }
                if (m_regression) {
                    DoubleValue dc = (DoubleValue) row.getCell(i);
                    output[0] = dc.getDoubleValue();
                } else {
                    for (int j = 0; j < nrOutputs; j++) {
                        if (classMap.get(row.getCell(i)) == j) {
                            output[j] = new Double(1.0);
                        } else {
                            output[j] = new Double(0.0);
                        }
                    }
                }
            }
        }
        if (add) {
            samples.add(sample);
            outputs.add(output);
            rowcounter++;
        }
    }
    Double[][] samplesarr = new Double[rowcounter][nrInputs];
    Double[][] outputsarr = new Double[rowcounter][nrInputs];
    for (int i = 0; i < samplesarr.length; i++) {
        samplesarr[i] = samples.get(i);
        outputsarr[i] = outputs.get(i);
    }
    // Now finally train the network.
    m_mlp.setClassMapping(classMap);
    m_mlp.setInputMapping(inputmap);
    RProp myrprop = new RProp();
    m_errors = new double[m_nrIterations.getIntValue()];
    for (int iteration = 0; iteration < m_nrIterations.getIntValue(); iteration++) {
        exec.setProgress((double) iteration / (double) m_nrIterations.getIntValue(), "Iteration " + iteration);
        myrprop.train(m_mlp, samplesarr, outputsarr);
        double error = 0;
        for (int j = 0; j < outputsarr.length; j++) {
            double[] myoutput = m_mlp.output(samplesarr[j]);
            for (int o = 0; o < outputsarr[0].length; o++) {
                error += (myoutput[o] - outputsarr[j][o]) * (myoutput[o] - outputsarr[j][o]);
            }
        }
        m_errors[iteration] = error;
        exec.checkCanceled();
    }
    // handle the optional PMML input
    PMMLPortObject inPMMLPort = m_pmmlInEnabled ? (PMMLPortObject) inData[INMODEL] : null;
    PMMLPortObjectSpec inPMMLSpec = null;
    if (inPMMLPort != null) {
        inPMMLSpec = inPMMLPort.getSpec();
    }
    PMMLPortObjectSpec outPortSpec = createPMMLPortObjectSpec(inPMMLSpec, posSpec, learningCols, targetCols);
    PMMLPortObject outPMMLPort = new PMMLPortObject(outPortSpec, inPMMLPort, posSpec);
    outPMMLPort.addModelTranslater(new PMMLNeuralNetworkTranslator(m_mlp));
    return new PortObject[] { outPMMLPort };
}

Example 10 with MultiLayerPerceptron

use of org.knime.base.data.neural.MultiLayerPerceptron in project knime-core by knime.

the class PMMLNeuralNetworkTranslator method addOutputLayer.

/**
 * Writes the PMML output layer of the MLP.
 *
 * @param nnModel
 *            the neural network model.
 * @param mlp
 *            the underlying {@link MultiLayerPerceptron}.
 * @param spec
 *            the port object spec
 */
protected void addOutputLayer(final NeuralNetwork nnModel, final MultiLayerPerceptron mlp, final PMMLPortObjectSpec spec) {
    int lastlayer = mlp.getNrLayers() - 1;
    String targetCol = spec.getTargetFields().iterator().next();
    Layer outputlayer = mlp.getLayer(lastlayer);
    Perceptron[] outputperceptrons = outputlayer.getPerceptrons();
    HashMap<DataCell, Integer> outputmap = mlp.getClassMapping();
    NeuralOutputs neuralOuts = nnModel.addNewNeuralOutputs();
    neuralOuts.setNumberOfOutputs(BigInteger.valueOf(outputperceptrons.length));
    for (int i = 0; i < outputperceptrons.length; i++) {
        NeuralOutput neuralOutput = neuralOuts.addNewNeuralOutput();
        neuralOutput.setOutputNeuron(lastlayer + "," + i);
        // search corresponding output value
        String colname = "";
        for (Entry<DataCell, Integer> e : outputmap.entrySet()) {
            if (e.getValue().equals(i)) {
                colname = ((StringValue) e.getKey()).getStringValue();
            }
        }
        DerivedField df = neuralOutput.addNewDerivedField();
        df.setOptype(OPTYPE.CATEGORICAL);
        df.setDataType(DATATYPE.STRING);
        if (mlp.getMode() == MultiLayerPerceptron.CLASSIFICATION_MODE) {
            df.setOptype(OPTYPE.CATEGORICAL);
            df.setDataType(DATATYPE.STRING);
        } else if (mlp.getMode() == MultiLayerPerceptron.REGRESSION_MODE) {
            df.setOptype(OPTYPE.CONTINUOUS);
            df.setDataType(DATATYPE.DOUBLE);
        }
        if (mlp.getMode() == MultiLayerPerceptron.CLASSIFICATION_MODE) {
            NormDiscrete normDiscrete = df.addNewNormDiscrete();
            normDiscrete.setField(targetCol);
            normDiscrete.setValue(colname);
        } else if (mlp.getMode() == MultiLayerPerceptron.REGRESSION_MODE) {
            FieldRef fieldRef = df.addNewFieldRef();
            fieldRef.setField(targetCol);
        }
    }
}