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

   * @return a value of type 'String'
   */
  public String header(String comparisonField) {

    if (!m_ResultsetsValid) {
      try {
	prepareData();
      } catch (Exception ex) {
	ex.printStackTrace();
	return ex.getMessage();
      }
    }
    return "Analysing:  "
      + comparisonField + '\n'
      + "Datasets:   " + getNumDatasets() + '\n'
      + "Resultsets: " + getNumResultsets() + '\n'
      + "Confidence: " + getSignificanceLevel() + " (two tailed)\n"
      + "Date:       " + (new SimpleDateFormat()).format(new Date()) + "\n\n";
  }


  /**
   * Carries out a comparison between all resultsets, counting the number
   * of datsets where one resultset outperforms the other.
   *
   * @param comparisonColumn the index of the comparison column
   * @return a 2d array where element [i][j] is the number of times resultset
   * j performed significantly better than resultset i.
   * @exception Exception if an error occurs
   */
  public int [][] multiResultsetWins(int comparisonColumn)
    throws Exception {

    int numResultsets = getNumResultsets();
    int [][] win = new int [numResultsets][numResultsets];
    for (int i = 0; i < numResultsets; i++) {
      for (int j = i + 1; j < numResultsets; j++) {
	System.err.print("Comparing (" + (i + 1) + ") with ("
			 + (j + 1) + ")\r");
	System.err.flush();
	for (int k = 0; k < getNumDatasets(); k++) {
	  try {
	    PairedStats pairedStats = 
	      calculateStatistics(m_DatasetSpecifiers.specifier(k), i, j,
				  comparisonColumn);
	    if (pairedStats.differencesSignificance < 0) {
	      win[i][j]++;
	    } else if (pairedStats.differencesSignificance > 0) {
	      win[j][i]++;
	    }
	  } catch (Exception ex) {
	    ex.printStackTrace();
	    System.err.println(ex.getMessage());
	  }
	}
      }
    }
    return win;
  }
  
  /**
   * Carries out a comparison between all resultsets, counting the number
   * of datsets where one resultset outperforms the other. The results
   * are summarized in a table.
   *
   * @param comparisonColumn the index of the comparison column
   * @return the results in a string
   * @exception Exception if an error occurs
   */
  public String multiResultsetSummary(int comparisonColumn)
    throws Exception {
    
    int [][] win = multiResultsetWins(comparisonColumn);
    int numResultsets = getNumResultsets();
    int resultsetLength = 1 + Math.max((int)(Math.log(numResultsets)
					     / Math.log(10)),
				       (int)(Math.log(getNumDatasets()) / 
					     Math.log(10)));
    String result = "";
    String titles = "";

    if (m_latexOutput) {
      result += "\\begin{table}[thb]\n\\caption{\\label{labelname}"
		  +"Table Caption}\n";
      result += "\\footnotesize\n";
      result += "{\\centering \\begin{tabular}{l";
    }

    for (int i = 0; i < numResultsets; i++) {
      if (m_latexOutput) {
	titles += " &";
	result += "c";
      }
      titles += ' ' + Utils.padLeft("" + (char)((int)'a' + i % 26),
				    resultsetLength);
    }
    if (m_latexOutput) {
      result += "}}\\\\\n\\hline\n";
      result += titles + " \\\\\n\\hline\n";
    } else {
      result += titles + "  (No. of datasets where [col] >> [row])\n";
    }
    for (int i = 0; i < numResultsets; i++) {
      for (int j = 0; j < numResultsets; j++) {
	if (m_latexOutput && j == 0) {
	  result +=  (char)((int)'a' + i % 26);
	}
	if (j == i) {
	  if (m_latexOutput) {
	    result += " & - ";
	  } else {
	    result += ' ' + Utils.padLeft("-", resultsetLength);
	  }
	} else {
	  if (m_latexOutput) {
	    result += "& " + win[i][j] + ' ';
	  } else {
	    result += ' ' + Utils.padLeft("" + win[i][j], resultsetLength);
	  }
	}
      }
      if (!m_latexOutput) {
	result += " | " + (char)((int)'a' + i % 26)
	  + " = " + getResultsetName(i) + '\n';
      } else {
	result += "\\\\\n";
      }
    }

    if (m_latexOutput) {
      result += "\\hline\n\\end{tabular} \\footnotesize \\par}\n\\end{table}";
    }
    return result;
  }

  public String multiResultsetRanking(int comparisonColumn)
    throws Exception {

    int [][] win = multiResultsetWins(comparisonColumn);
    int numResultsets = getNumResultsets();
    int [] wins = new int [numResultsets];
    int [] losses = new int [numResultsets];
    int [] diff = new int [numResultsets];
    for (int i = 0; i < win.length; i++) {
      for (int j = 0; j < win[i].length; j++) {
	wins[j] += win[i][j];
	diff[j] += win[i][j];
	losses[i] += win[i][j];
	diff[i] -= win[i][j];
      }
    }
    int biggest = Math.max(wins[Utils.maxIndex(wins)],
			   losses[Utils.maxIndex(losses)]);
    int width = Math.max(2 + (int)(Math.log(biggest) / Math.log(10)),
			 ">-<".length());
    String result;
    if (m_latexOutput) {
      result = "\\begin{table}[thb]\n\\caption{\\label{labelname}Table Caption"
	+"}\n\\footnotesize\n{\\centering \\begin{tabular}{rlll}\\\\\n\\hline\n";
      result += "Resultset & Wins$-$ & Wins & Losses \\\\\n& Losses & & "
	+"\\\\\n\\hline\n";
    } else {
      result = Utils.padLeft(">-<", width) + ' '
	+ Utils.padLeft(">", width) + ' '
	+ Utils.padLeft("<", width) + " Resultset\n";
    }
    int [] ranking = Utils.sort(diff);
    for (int i = numResultsets - 1; i >= 0; i--) {
      int curr = ranking[i];
      if (m_latexOutput) {
	result += "(" + (curr+1) + ") & " 
	  + Utils.padLeft("" + diff[curr], width) 
	  +" & " + Utils.padLeft("" + wins[curr], width)
	  +" & " + Utils.padLeft("" + losses[curr], width)
	  +"\\\\\n";
      } else {
	result += Utils.padLeft("" + diff[curr], width) + ' '
	  + Utils.padLeft("" + wins[curr], width) + ' '
	  + Utils.padLeft("" + losses[curr], width) + ' '
	  + getResultsetName(curr) + '\n';
      }
    }

    if (m_latexOutput) {
      result += "\\hline\n\\end{tabular} \\footnotesize \\par}\n\\end{table}";
    }
    return result;
  }

  /**
   * Generates a comparison table in latex table format
   *
   * @param baseResultset the index of the base resultset
   * @param comparisonColumn the index of the column to compare over
   * @param maxWidthMean width for the mean
   * @param maxWidthStdDev width for the standard deviation
   * @return the comparison table string
   */
  private String multiResultsetFullLatex(int baseResultset,
				     int comparisonColumn,
				     int maxWidthMean,
				     int maxWidthStdDev) {

    StringBuffer result = new StringBuffer(1000);
    int numcols = getNumResultsets() * 2;
    if (m_ShowStdDevs) {
      numcols += getNumResultsets();
    }

    result.append("\\begin{table}[thb]\n\\caption{\\label{labelname}"
		  +"Table Caption}\n");
    if (!m_ShowStdDevs) {
      result.append("\\footnotesize\n");
    } else {
      result.append("\\scriptsize\n");
    }

    // output the column alignment characters
    // one for the dataset name and one for the comparison column
    if (!m_ShowStdDevs) {
      result.append("{\\centering \\begin{tabular}{ll");
    } else {
      // dataset, mean, std dev
      result.append("{\\centering \\begin{tabular}{lr@{\\hspace{0cm}}l");
    }

    for (int j = 0; j < getNumResultsets(); j++) {
      if (j != baseResultset) {
	if (!m_ShowStdDevs) {
	  result.append("l@{\\hspace{0.1cm}}l");
	} else {
	  result.append("r@{\\hspace{0cm}}l@{\\hspace{0cm}}r");
	}
      }
    }
    result.append("}\n\\\\\n\\hline\n");
    if (!m_ShowStdDevs) {
      result.append("Data Set & ("+(baseResultset+1)+")");
    } else {
      result.append("Data Set & \\multicolumn{2}{c}{("+(baseResultset+1)+")}");
    }

    // now do the column names (numbers)
    for (int j = 0; j < getNumResultsets(); j++) {
      if (j != baseResultset) {
	if (!m_ShowStdDevs) {
	  result.append("& (" + (j + 1) + ") & ");
	} else {
	  result.append("& \\multicolumn{3}{c}{(" + (j + 1) + ")} ");
	}
      }
    }
    result.append("\\\\\n\\hline\n");
    
    int datasetLength = 25;
    int resultsetLength = maxWidthMean + 7;
    if (m_ShowStdDevs) {
      resultsetLength += (maxWidthStdDev + 5);
    }

    for (int i = 0; i < getNumDatasets(); i++) {
      // Print the name of the dataset
      String datasetName = 
	templateString(m_DatasetSpecifiers.specifier(i)).replace('_','-');
      try {
	PairedStats pairedStats = 
	  calculateStatistics(m_DatasetSpecifiers.specifier(i), 
			      baseResultset, baseResultset,
			      comparisonColumn);
	datasetName = Utils.padRight(datasetName, datasetLength);
	result.append(datasetName);

	
	if (!m_ShowStdDevs) {
	  result.append("& "+Utils.doubleToString(pairedStats.xStats.mean,
			       resultsetLength - 2, 2));
	} else {
	  result.append("& "+Utils.doubleToString(pairedStats.xStats.mean,
					     (maxWidthMean+5), 2)+"$\\pm$");
	  if (Double.isNaN(pairedStats.xStats.stdDev)) {
	    result.append("&"+Utils.doubleToString(0.0,
						  (maxWidthStdDev+3),2)+" ");
	  } else {
	    result.append("&"+Utils.doubleToString(pairedStats.xStats.stdDev,
						   (maxWidthStdDev+3),2)+" ");
	  }
	}
	// Iterate over the resultsets
	for (int j = 0; j < getNumResultsets(); j++) {
	  if (j != baseResultset) {
	    try {
	      pairedStats = 
		calculateStatistics(m_DatasetSpecifiers.specifier(i), 
				    baseResultset, j, comparisonColumn);
	      String sigString = "";
	      if (pairedStats.differencesSignificance < 0) {
		sigString = "$\\circ$";
	      } else if (pairedStats.differencesSignificance > 0) {
		sigString = "$\\bullet$";
	      } 
	      if (!m_ShowStdDevs) {
		result.append(" & "+Utils.doubleToString(pairedStats.yStats.mean,
						   resultsetLength - 2,
						   2)).append(" & "+sigString);
	      } else {
		result.append(" & "
			      +Utils.doubleToString(pairedStats.yStats.mean,
						   (maxWidthMean+5),
						   2)+"$\\pm$");
		if (Double.isNaN(pairedStats.yStats.stdDev)) {
		  result.append("&"+Utils.doubleToString(0.0, 
				(maxWidthStdDev+3),2)+" ");
		} else {
		  result.append("&"+Utils.doubleToString(pairedStats.
				  yStats.stdDev, (maxWidthStdDev+3),2)+" ");
		}
		result.append(" & ").append(sigString);
	      }
	    } catch (Exception ex) {
	      ex.printStackTrace();
	      result.append(Utils.padLeft("", resultsetLength + 1));
	    }
	  }
	}
	result.append("\\\\\n");
      } catch (Exception ex) {
	ex.printStackTrace();
      }
    }

    result.append("\\hline\n\\multicolumn{"+numcols+"}{c}{$\\circ$, $\\bullet$"
		  +" statistically significant improvement or degradation}"
		  +"\\\\\n\\end{tabular} ");
    if (!m_ShowStdDevs) {
      result.append("\\footnotesize ");
      } else {
	result.append("\\scriptsize ");
      }
    
    result.append("\\par}\n\\end{table}"
		  +"\n");
    System.out.println(result.toString()+"\n\n");
    return result.toString();
  }


  /**
   * Generates a comparison table in latex table format
   *
   * @param baseResultset the index of the base resultset
   * @param comparisonColumn the index of the column to compare over
   * @param maxWidthMean width for the mean
   * @param maxWidthStdDev width for the standard deviation
   * @return the comparison table string
   */
  private String multiResultsetFullPlainText(int baseResultset,
                                             int comparisonColumn,
                                             int maxWidthMean,
                                             int maxWidthStdDev) {

    StringBuffer result = new StringBuffer(1000);
    int datasetLength = 25;
    //=============== BEGIN EDIT melville ===============
    int meanWidth = maxWidthMean + 2;
    int meanWidthStdDev = maxWidthStdDev;
    if(m_Precision>0) {//account for the decimal point and decimal digits
	meanWidth += (1 + m_Precision); 
	meanWidthStdDev += (1 + m_Precision);
    }
    int resultsetLength = meanWidth + 2;
    if (m_ShowStdDevs) {
	resultsetLength += (meanWidthStdDev + 2);
    }
    //=============== END EDIT melville ===============
    
    // Set up the titles
    StringBuffer titles = new StringBuffer(Utils.padRight("Dataset",
                                                          datasetLength));
    titles.append(' ');
    StringBuffer label 
      = new StringBuffer(Utils.padLeft("(" + (baseResultset + 1)
                                       + ") "
                                       + getResultsetName(baseResultset),
                                       resultsetLength + 3));

    titles.append(label);
    StringBuffer separator = new StringBuffer(Utils.padRight("",
                                                             datasetLength));
    while (separator.length() < titles.length()) {
      separator.append('-');
    }
    separator.append("---");
    titles.append(" | ");