pairedttester.java

来自「wekaUT是 university texas austin 开发的基于wek」· Java 代码 · 共 1,963 行 · 第 1/5 页

								      comparisonColumn);
			result.append(Utils.padLeft("",4));
			result.append(Utils.doubleToString(pairedStats.xStats.mean,
							   meanWidth, m_Precision));
			if (m_ShowStdDevs) {
			    if (Double.isInfinite(pairedStats.xStats.stdDev)) {
				result.append('(' + Utils.padRight("Inf", stdWidth)
					      +')');
			    } else {
				result.append('('+Utils.doubleToString(pairedStats.xStats.stdDev,
								       stdWidth,m_Precision)
					      +')');
			    }
			}
		    }catch (Exception ex) {
			ex.printStackTrace();
		    }
		}else{//if the point was not tested print spaces
		    result.append(Utils.padLeft("",resultsetLength));
		}
	    }
	    result.append("\n");
	}
	result.append("\n\n");
	return result.toString();
    }
    //=============== END EDIT melville ===============


    //=============== BEGIN EDIT melville ===============
  /**
   * Creates a comparison table where a base resultset is compared to the
   * other resultsets. Results are presented for every dataset.
   *
   * @param baseResultset the index of the base resultset
   * @param comparisonColumn the index of the column to compare over
   * @return the comparison table string
   * @exception Exception if an error occurs
   */
    public String multiResultsetPercentErrorReduction(int baseResultset,
							  int comparisonColumn) throws Exception {

	StringBuffer result = new StringBuffer(1000);
        if(getNumResultsets()==1){//no comparison to be made - just display one learning curve
	    int maxWidthMean = 2;
	    int maxWidthStdDev = 2;
	    // determine max field width
	    for (int i = 0; i < getNumDatasets(); i++) {
		for (int j = 0; j < getNumResultsets(); j++) {
		    try {
			PairedStats pairedStats = 
			    calculateStatistics(m_DatasetSpecifiers.specifier(i), 
						baseResultset, j, comparisonColumn);
			if (!Double.isInfinite(pairedStats.yStats.mean) &&
			    !Double.isNaN(pairedStats.yStats.mean)) {
			    double width = ((Math.log(Math.abs(pairedStats.yStats.mean)) / 
					     Math.log(10))+1);
			    if (width > maxWidthMean) {
				maxWidthMean = (int)width;
			    }
			}
			if (m_ShowStdDevs &&
			    !Double.isInfinite(pairedStats.yStats.stdDev) &&
			    !Double.isNaN(pairedStats.yStats.stdDev)) {
			    double width = ((Math.log(Math.abs(pairedStats.yStats.stdDev)) / 
					     Math.log(10))+1);
			    if (width > maxWidthStdDev) {
				maxWidthStdDev = (int)width;
			    }
			}
		    }  catch (Exception ex) {
			ex.printStackTrace();
		    }
		}
	    }
	    result = new StringBuffer(oneResultsetFullLearningPlainText(baseResultset, 
									comparisonColumn, 
									maxWidthMean,
									maxWidthStdDev));
	}
	else
	    result = new StringBuffer(multiResultsetPercentErrorReductionPlainText(baseResultset, 
										   comparisonColumn));
    
	// append a key so that we can tell the difference between long
	// scheme+option names
	result.append("\nKey:\n\n");
	for (int j = 0; j < getNumResultsets(); j++) {
	    result.append("("+(j+1)+") ");
	    result.append(getResultsetName(j)+"\n");
	}
	return result.toString();
    }

//=============== END EDIT melville ===============



    //=============== BEGIN EDIT melville ===============
    /**
     * Generates percentage error reduction tables across learning curves
     *
     * @param baseResultset the index of the base resultset
     * @param comparisonColumn the index of the column to compare over
     * @return the comparison table string
     */
    private String multiResultsetPercentErrorReductionPlainText(int baseResultset,
								int comparisonColumn) {
	
	StringBuffer result = new StringBuffer(1000);
	//Compute column widths for pretty printing
	int datasetLength = 20;
	int meanWidth = 3; 
	if(m_Precision>0) meanWidth += (1 + m_Precision); //account for the decimal point and decimal digits
	int resultsetLength = 2*meanWidth;
	int maxPointWidth = 6;
	int numPts = m_Points.length;
	
	//Find maximum point width - assuming the points are in order
	maxPointWidth = (int) (Math.log(m_Points[m_Points.length-1])/Math.log(10)) + 1;
	if(maxPointWidth<6) maxPointWidth=6;
	
	Vector datasetNames = new Vector();//store names of dataset
	HashMap specMap = new HashMap();//maps dataset:point pair to a unique specifier
	
	for (int i = 0; i < getNumDatasets(); i++) {
	    Instance spec = m_DatasetSpecifiers.specifier(i);
	    String dataName = spec.toString(m_DatasetKeyColumns[0]);
	    String point = spec.toString(m_DatasetKeyColumns[1]);
	    if(!datasetNames.contains(dataName)){
		datasetNames.add(dataName);
	    }
	    specMap.put(dataName+":"+point, spec);
	}
		
	StringBuffer titles;
	for (int j = 0; j < getNumResultsets(); j++) {//For each system comparison
	    //Generate a different table comparing pts across the learning curve with the base system
	    
	    if (j == baseResultset) continue;
	    
	    //Display title
	    titles = new StringBuffer();
	    titles.append("Comparing "+Utils.padLeft("(" + (j + 1)+ ") "+ getResultsetName(j),
						     15)+" to "+
			  Utils.padRight("(" + (baseResultset + 1)+ ") "+ getResultsetName(baseResultset),
					 15));
	    result.append(titles.toString()).append("\n\n");
	    
	    //Display table headings
	    result.append("Dataset\n");
	    if(m_Fraction)
		result.append(Utils.padLeft("% Points",datasetLength));
	    else
		result.append(Utils.padLeft("Points",datasetLength));

	    /****************************************
	     * Uncomment out the block below, if you want the error
	     * reductions displayed for each point on the curve.
	     ****************************************/

	    //  	    for(int pts=0; pts<numPts; pts++){
	    //  		if(m_Fraction)
	    //  		    result.append(Utils.padLeft(Utils.doubleToString(m_Points[pts]*100,meanWidth,2),resultsetLength));
	    //  		else
	    //  		    result.append(Utils.padLeft(Utils.doubleToString(m_Points[pts],meanWidth,2),resultsetLength));
	    //  	    }
	    //  	    result.append("\n");
	    //  	    result.append(Utils.padLeft("",datasetLength));
	    //  	    for(int pts=0; pts<numPts; pts++)
	    //  		for(int k=0; k<resultsetLength; k++)
	    //  		    result.append("-");
  	    result.append("\n");
	    
	    int datasetCtr = 0;
	    double errorReduction;
	    double datasetSum = 0.0;//sum of avg error reduction across datasets
	    for(int dataIndex=0; dataIndex<datasetNames.size(); dataIndex++){//for each dataset
		double r_sum = 0.0;//sum across the row
		int r_ctr = 0;//num of pts in the row (this can be less than numPts)

		result.append(Utils.padRight((String)datasetNames.get(dataIndex), datasetLength));
		
		for(int ptIndex=0; ptIndex<m_Points.length; ptIndex++){//for each point on the curve
		    String key = datasetNames.get(dataIndex)+":"+(new Double(m_Points[ptIndex]));
		    Object obj = specMap.get(key);
		    if(!m_Fraction && obj==null){ 
			key = datasetNames.get(dataIndex)+":"+(new Integer((int) m_Points[ptIndex]));
			obj = specMap.get(key);
		    }
		    if(obj!=null){//if the point was recorded
			Instance specifier = (Instance) obj;
			try {
			    PairedStats pairedStats =  calculateStatistics(specifier, baseResultset, j, comparisonColumn);
			    double yVal = pairedStats.yStats.mean;
			    pairedStats = calculateStatistics(specifier, baseResultset, baseResultset, comparisonColumn);
			    double xVal = pairedStats.xStats.mean;
			    if(yVal == 0 && xVal == 0)
				errorReduction = 0;
			    else
				errorReduction = (1.0 - yVal/xVal)*100.0;
			    //Check if errorReduction is NaN or Inf - else
			    //result.append(Utils.padLeft(Utils.doubleToString(errorReduction,meanWidth,m_Precision),resultsetLength));
			    r_sum += errorReduction;
			    r_ctr++;
			}catch (Exception ex) {
			    ex.printStackTrace();
			}
		    }else{//if the point was not tested print spaces
			//result.append(Utils.padLeft("",resultsetLength));
		    }
		}
		if(r_ctr>0){
		    datasetCtr++;
		    double r_avg = r_sum/r_ctr;
		    datasetSum += r_avg;
		    result.append(" | "+Utils.padLeft(Utils.doubleToString(r_avg,meanWidth,m_Precision),resultsetLength-3));
		}
		    result.append("\n");
	    }
	    String s="";
	    for(int k=0; k<meanWidth; k++)
		s = s + "-";
	    /****************************************
	     * Uncomment out the block below, if you want the error
	     * reductions displayed for each point on the curve.
	     ****************************************/
	    //  	    result.append(Utils.padLeft(s,datasetLength+resultsetLength*(numPts+1)));
	    //  	    result.append("\n");
	    //  	    result.append(Utils.padLeft(Utils.doubleToString(datasetSum/datasetCtr,meanWidth,m_Precision),datasetLength+resultsetLength*(numPts+1))+"\n\n\n");
	    result.append(Utils.padLeft(s,datasetLength+resultsetLength));
	    result.append("\n");
	    result.append(Utils.padLeft(Utils.doubleToString(datasetSum/datasetCtr,meanWidth,m_Precision),datasetLength+resultsetLength)+"\n\n\n");
	}
    	return result.toString();
    }
    //=============== END EDIT melville ===============

	
    //=============== BEGIN EDIT melville ===============
  /**
   * Creates a comparison table where a base resultset is compared to the
   * other resultsets. Results are presented for every dataset.
   *
   * @param baseResultset the index of the base resultset
   * @param comparisonColumn the index of the column to compare over
   * @param nFraction top n fraction of error reductions are averaged
   * @return the comparison table string
   * @exception Exception if an error occurs */
    public String multiResultsetTopNPercentErrorReduction(int baseResultset,
							  int comparisonColumn, double nFraction) throws Exception {
	
	StringBuffer result = new StringBuffer(1000);
        if(getNumResultsets()==1){//no comparison to be made - just display one learning curve
	    int maxWidthMean = 2;
	    int maxWidthStdDev = 2;
	    // determine max field width
	    for (int i = 0; i < getNumDatasets(); i++) {
		for (int j = 0; j < getNumResultsets(); j++) {
		    try {
			PairedStats pairedStats = 
			    calculateStatistics(m_DatasetSpecifiers.specifier(i), 
						baseResultset, j, comparisonColumn);
			if (!Double.isInfinite(pairedStats.yStats.mean) &&
			    !Double.isNaN(pairedStats.yStats.mean)) {
			    double width = ((Math.log(Math.abs(pairedStats.yStats.mean)) / 
					     Math.log(10))+1);
			    if (width > maxWidthMean) {
				maxWidthMean = (int)width;
			    }
			}
			if (m_ShowStdDevs &&
			    !Double.isInfinite(pairedStats.yStats.stdDev) &&
			    !Double.isNaN(pairedStats.yStats.stdDev)) {
			    double width = ((Math.log(Math.abs(pairedStats.yStats.stdDev)) / 
					     Math.log(10))+1);
			    if (width > maxWidthStdDev) {
				maxWidthStdDev = (int)width;
			    }
			}
		    }  catch (Exception ex) {
			ex.printStackTrace();
		    }
		}
	    }
	    result = new StringBuffer(oneResultsetFullLearningPlainText(baseResultset, 
									comparisonColumn, 
									maxWidthMean,
									maxWidthStdDev));
	}
	else{
	    result = new StringBuffer(multiResultsetTopNPercentErrorReductionPlainText(baseResultset, 
										       comparisonColumn, nFraction));
	}
	
	// append a key so that we can tell the difference between long
	// scheme+option names
	result.append("\nKey:\n\n");
	for (int j = 0; j < getNumResultsets(); j++) {
	    result.append("("+(j+1)+") ");
	    result.append(getResultsetName(j)+"\n");
	}
	return result.toString();
    }

//=============== END EDIT melville ===============


    //=============== BEGIN EDIT melville ===============
    /**
     * Generates percentage error reduction tables across learning curves
     *
     * @param baseResultset the index of the base resultset
     * @param comparisonColumn the index of the column to compare over
     * @param nFraction top n fraction of error reductions are averaged
     * @return the comparison table string
     */
    private String multiResultsetTopNPercentErrorReductionPlainText(int baseResultset,
								    int comparisonColumn, double nFraction) {
	
	StringBuffer result = new StringBuffer(1000);
	//Compute column widths for pretty printing
	int datasetLength = 20;
	int meanWidth = 3; 
	if(m_Precision>0) meanWidth += (1 + m_Precision); //account for the decimal point and decimal digits
	int resultsetLength = 2*meanWidth;
	int maxPointWidth = 6;
	int numPts = m_Points.length;
	
	//Find maximum point width - assuming the points are in order
	maxPointWidth = (int) (Math.log(m_Points[m_Points.length-1])/Math.log(10)) + 1;
	if(maxPointWidth<6) maxPointWidth=6;
	
	Vector datasetNames = new Vector();//store names of dataset
	HashMap specMap = new HashMap();//maps dataset:point pair to a unique specifier
	
	for (int i = 0; i < getNumDatasets(); i++) {
	    Instance spec = m_DatasetSpecifiers.specifier(i);
	    String dataName = spec.toString(m_DatasetKeyColumns[0]);
	    String point = spec.toString(m_DatasetKeyColumns[1]);
	    if(!datasetNames.contains(dataName)){
		datasetNames.add(dataName);
	    }
	    specMap.put(dataName+":"+point, spec);
	}
	
	StringBuffer titles;
	for (int j = 0; j < getNumResultsets(); j++) {//For each system comparison
	    //Generate a different table comparing pts across the learning curve with the base system
	    
	    if (j == baseResultset) continue;
	    
	    //Display title
	    titles = new StringBuffer();
	    titles.append("Comparing "+Utils.padLeft("(" + (j + 1)+ ") "+ getResultsetName(j),
						     15)+" to "+
			  Utils.padRight("(" + (baseResultset + 1)+ ") "+ getResultsetName(baseResultset),
					 15));
	    result.append(titles.toString()).append("\n\n");
	    
	    //Display table headings
	    result.append("Dataset\n");
	    if(m_Fraction)
		result.append(Utils.padLeft("% Points",datasetLength));
	    else
		result.append(Utils.padLeft("Points",datasetLength));

	    /****************************************
	     * Uncomment out the block below, if you want the error
	     * reductions displayed for each point on the curve.
	     ****************************************/	    
	    //  	    for(int pts=0; pts<numPts; pts++){
	    //  		if(m_Fraction)
	    //  		    result.append(Utils.padLeft(Utils.doubleToString(m_Points[pts]*100,meanWidth,2),resultsetLength));
	    //  		else
	    //  		    result.append(Utils.padLeft(Utils.doubleToString(m_Points[pts],meanWidth,2),resultsetLength));
	    //  	    }
	    //  	    result.append("\n");
	    //  	    result.append(Utils.padLeft("",datasetLength));
	    //  	    for(int pts=0; pts<numPts; pts++)
	    //  		for(int k=0; k<resultsetLength; k++)
	    //  		    result.append("-");
	    result.append("\n");
	    
	    int datasetCtr = 0;
	    double errorReduction;
	    
	    Triple []ptsErrorReduction = new Triple[numPts];//stores error reductions at each point and t-test results
	    double datasetSum = 0.0;//sum of avg error reduction across datasets
	    for(int dataIndex=0; dataIndex<datasetNames.size(); dataIndex++){//for each dataset
		int r_ctr = 0;