attributedelegate.java

Fast implementation of C4/5 in Java

/**
 * @(#)AttributeDelegate.java      1.5.0 09/01/18
 */

package ml.classifier.dt;

/**
 * A delegate of an attribute, containing some essential processed
 * information about the attribute to speed up the tree building process.
 *
 * @author Ping He
 * @author Xiaohua Xu
 * @see ml.classifier.dt.DiscreteAttributeDelegate
 * @see ml.classifier.dt.ContinuousAttributeDelegate
 */
public abstract class AttributeDelegate{
	/**
	 * Whether there is any missing data on the attribute
	 */
	protected boolean hasMissingData;

 	/**
 	 * Initialize an attribute delegate
 	 */
	protected AttributeDelegate(){
		// Before we find missing data, it is false
		this.hasMissingData = false;
	}

	/**
	 * Retrieve whether there is any missing data on the attribute
	 */
	public boolean hasMissingData(){
		return hasMissingData;
	}

	/**
	 * Set whether there is any missing data on the attribute
	 */
	public void setHasMissingData(boolean value){
		hasMissingData = value;
	}

	/**
	 * Set the original sequence and weight for each data.
	 */
	public abstract void setCasesWeight(int[] cases, float[] weight);

	/**
	 * Evaluate the Gain and splitInfo value for the specified data when it splits on
	 * the attribute.
	 *
	 * @param first the begin (inclusive) index of the data to be evaluated
	 * @param last the end (exclusive) index of the data to be evaluated
	 * @param classAttributeDelegate the delegate of the class attribute. It helps when computing
	 *         data's class distribution.
	 * @return The evaluation result
	 *
	 * @see ml.classifier.dt.GainCalculator
	 */
	protected abstract float[] evaluate(int first, int last, AttributeDelegate classAttributeDelegate);

	/**
	 * Group the data with the specified branch value forward and compute its branch weight.
	 *
	 * @param first the begin (inclusive) index of the data to be grouped
	 * @param last the end (exclusive) index of the data to be grouped
	 * @param groupBranch For discrete attribute, the branch index to be grouped;
	 *                    For continuous attribute, -1 for missing data,
	 *                                              otherwise the rank of the cut value.
	 * @param branchDistri Actually an output of this method, recording the weight of each branch.
	 *        
	 * @return The boundary index before which the specified data is grouped
	 */
	protected abstract int groupForward(int first, int last, int groupBranch, float[] branchDistri);

	/**
	 * Group the data with missing value on the attribute backward. 
	 * <p>
	 * The reason for grouping missing data backward is to narrow the grouping range for the next branch.
	 * </p>
	 * @param first the begin (inclusive) index of the data to be grouped
	 * @param last the end (exclusive) index of the data to be grouped
	 * @return The boundary index after which the data is grouped
	 *
	 */
	protected abstract int groupBackward(int first, int last);

	/**
	 * Get the number of branches if the attribute is selected as the test attribute.
	 * <p>
	 * When an attribute is selected as the test attribute, missing value is not taken as
	 * a valid branch value.
	 * </p>
	 */
	protected abstract int getBranchCount();

	/**
	 * Get the branch index of the class attribute value of the specified data. 
	 * <p>
	 * Only supported by discrete attribute delegates. <br>
	 * Only used by class attribute delegate.
	 * </p>
	 *
	 * @param caseIndex the index of the specified data
	 * @return The branch index of the class value of the specified data.
	 * @see ml.classifier.dt.DiscreteAttributeDelegate#getClassBranch(int caseIndex)
	 */
	protected int getClassBranch(int caseIndex){
		throw new UnsupportedOperationException("Only Supported By The Class Attribute!");
	}

	/**
	 * Find the rank of the cut value in the test attribute.
	 * <p>Only supported by continuous attribute delegates.</p>
	 *
	 * @param preSplitRank the begin (inclusive) rank from which the search of cut should start
	 * @param splitRank the end (inclusive) rank to which the search of cut should stop
	 * @return The rank of the cut value
     * @see ml.classifier.dt.ContinuousAttributeDelegate#findCutRank(int splitRank, int preSplitRank)
	 */
	protected int findCutRank(int splitRank, int preSplitRank) {
		throw new UnsupportedOperationException("Only Supported By Continuous Attribute!");
	}

	/**
	 * Find the cut value of the test attribute when provided with its rank.
	 * <p> Only supported by continuous attribute delegates.</p>
	 *
	 * @param cutRank the rank of the cut
	 * @return The cut value of the test attribute
	 *
	 * @see ml.classifier.dt.ContinuousAttributeDelegate#findCutRank(int splitRank, int preSplitRank)
	 */
	protected float findCut(int cutRank){
		throw new UnsupportedOperationException("Only Supported By Continuous Attribute!");
	}
}