neuralnetworkalgorithm.java

一个数据挖掘软件ALPHAMINERR的整个过程的JAVA版源代码

/*
 *    This program is free software; you can redistribute it and/or modify
 *    it under the terms of the GNU General Public License as published by
 *    the Free Software Foundation; either version 2 of the License, or
 *    (at your option) any later version.
 *
 *    This program is distributed in the hope that it will be useful,
 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *    GNU General Public License for more details.
 *
 *    You should have received a copy of the GNU General Public License
 *    along with this program; if not, write to the Free Software
 *    Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

/**
 * Title: XELOPES Data Mining Library
 * Description: The XELOPES library is an open platform-independent and data-source-independent library for Embedded Data Mining.
 * Copyright: Copyright (c) 2002 Prudential Systems Software GmbH
 * Company: ZSoft (www.zsoft.ru), Prudsys (www.prudsys.com)
 * @author Michael Thess
 * @version 1.2
 */
package com.prudsys.pdm.Models.Regression.NeuralNetwork;

import com.prudsys.pdm.Core.ApplicationAttribute;
import com.prudsys.pdm.Core.MiningException;
import com.prudsys.pdm.Core.MiningModel;
import com.prudsys.pdm.Core.MiningSettings;
import com.prudsys.pdm.Input.MiningArrayStream;
import com.prudsys.pdm.Models.Regression.RegressionAlgorithm;
import com.prudsys.pdm.Models.Supervised.Classifier;
import com.prudsys.pdm.Transform.MiningTransformationActivity;
import com.prudsys.pdm.Transform.Special.ReplaceMissingValueStream;
import com.prudsys.pdm.Transform.Special.TreatOutlierValueStream;

/**
 * A class representing a Neural Network algorithm. Each implementation
 * should extend this class and override only the methods:
 * {@link #runAlgorithm() runAlgorithm()},
 * {@link #getClassifier()() getClassifier()}
 */
public abstract class NeuralNetworkAlgorithm extends RegressionAlgorithm
{
    // -----------------------------------------------------------------------
    //  Variables declarations
    // -----------------------------------------------------------------------
    /** Type of learning algorithm. */
    protected int learningType = NeuralNetworkSettings.BACK_PROPAGATION_WITH_MOMENTUM;

    /** Algorithm automatically builds the neural network. */
    protected boolean autoBuildNetwork = true;

    /** Learning rate of backpropagation method. */
    protected double learningRate = 1.0;

    /** Momentum of backpropagation method, if used. */
    protected double momentum = 0.5;

    /** Maximum number of iterations. */
    protected int maxNumberOfIterations = 1000;

    /** Maximum acceptable error. */
    protected double maxError = 1.0E-35;

    /** Neural network object, if algorithm doesn't build it automatically. */
    protected NeuralNetwork neuralNetwork = null;

    // -----------------------------------------------------------------------
    //  Getter and setter methods
    // -----------------------------------------------------------------------
    /**
     * Creates an instance of the neural network settings class that is required
     * to run the algorithm. The mining settings are assigned through the
     * setMiningSettings method.
     *
     * @return new instance of the sparse grids settings class of the algorithm
     */
    public MiningSettings createMiningSettings() {

      return new NeuralNetworkSettings();
    }

    /**
     * Set NN settings.
     *
     * @param miningSettings instance of NeuralNetworkSettings
     * @exception IllegalArgumentException mining settings are not of Neural Network type
     */
    public final void setMiningSettings( MiningSettings miningSettings ) throws IllegalArgumentException
    {
        if ( miningSettings instanceof NeuralNetworkSettings )
        {
            super.setMiningSettings( miningSettings );
            NeuralNetworkSettings nns  = (NeuralNetworkSettings) miningSettings;
            this.learningType          = nns.getLearningType();
            this.autoBuildNetwork      = nns.isAutoBuildNetwork();
            this.learningRate          = nns.getLearningRate();
            this.momentum              = nns.getMomentum();
            this.maxNumberOfIterations = nns.getMaxNumberOfIterations();
            this.maxError              = nns.getMaxError();
            this.neuralNetwork         = nns.getNeuralNetwork();
        }
        else
        {
            throw new IllegalArgumentException( "MiningSettings have to be NeuralNetworkSettings." );
        };
    }

    /**
     * Returns NN classifier. By default, this is the neuralNetwork object.
     *
     * @return NN classifier
     */
    public Classifier getClassifier() {

      return neuralNetwork;
    }

    // -----------------------------------------------------------------------
    // Run Neural Network algorithm and build model
    // -----------------------------------------------------------------------
    /**
     * Builds mining model by running the NN algorithm internally.
     * Before starting the algorithm, missing values are replaced.
     *
     * @return sparse grids mining model generated by the algorithm
     * @exception MiningException could not build model
     */
    public MiningModel buildModel() throws MiningException
    {
      long start = ( new java.util.Date() ).getTime();

      // Outlier treatment and missing value replacement:
      TreatOutlierValueStream tro   = new TreatOutlierValueStream(miningInputStream);
      tro.setNumOutliers( ApplicationAttribute.OUTLIER_TREATMENT_METHOD_asExtremeValues );
      tro.createMiningTransformationStep();

      ReplaceMissingValueStream rep = new ReplaceMissingValueStream(miningInputStream);
      miningInputStream             = new MiningArrayStream( rep.createTransformedStream() );

      // Transform unbounded -> bounded categorical attributes:
      metaData.getMetaDataOp().unboundedToBoundedCategories();

      // Run NN algorithm:
      runAlgorithm();

      // Build NN model:
      NeuralNetworkModel model = new NeuralNetworkModel();
      model.setMiningSettings( miningSettings );

      model.setInputSpec( applicationInputSpecification );
      model.setTarget( applicationInputSpecification.getTargetApplicationAttribute() );

      // Outlier treatment and missing value in application input specification:
      // Create inner transformation object:
      MiningTransformationActivity mta = new MiningTransformationActivity();
      mta.addTransformationStep( tro.getMts() );
      mta.addTransformationStep( rep.getMts() );
      model.setMiningTransform( mta );

      // Outliers and missing values in application input specification:
      applicationInputSpecification.setInputSpecFromInnerTrafo(metaData, tro, rep);

      // No NN model parameter to set.

      // Set classifier:
      model.setClassifier( getClassifier() );

      this.miningModel = model;

      long end = ( new java.util.Date() ).getTime();
      timeSpentToBuildModel = ( end - start ) / 1000.0;

      return model;
    }

    /**
     * Runs NN algorithm.
     *
     * @exception MiningException could not run algorithm
     */
    protected abstract void runAlgorithm() throws MiningException;
}