在Eclipse中生成ROC曲线(weka)

时间:2017-03-19 08:40:53

标签: java eclipse classification weka roc

我使用以下代码来训练/测试一些分类器:

public class WekaTest {
  public static BufferedReader readDataFile(String filename) 
  {
    BufferedReader inputReader = null;
    try
    {
        inputReader = new BufferedReader(new FileReader(filename));
    }
    catch (FileNotFoundException ex) 
    {
        System.err.println("File not found: " + filename);
    }

    return inputReader;
}

public static Evaluation classify(Classifier model,
        Instances trainingSet, Instances testingSet) throws Exception {
    Evaluation evaluation = new Evaluation(trainingSet);

    model.buildClassifier(trainingSet);
    evaluation.evaluateModel(model, testingSet);

    return evaluation;
}

public static double calculateAccuracy(FastVector predictions) {
    double correct = 0;

    for (int i = 0; i < predictions.size(); i++) {
        NominalPrediction np = (NominalPrediction) predictions.elementAt(i);
        if (np.predicted() == np.actual()) {
            correct++;
        }
    }

    return 100 * correct / predictions.size();
}

public static Instances[][] crossValidationSplit(Instances data, int numberOfFolds) {
    Instances[][] split = new Instances[2][numberOfFolds];
    Random random = new Random();
    for (int i = 0; i < numberOfFolds; i++) 
    {
        split[0][i] = data.trainCV(numberOfFolds, i, random);
        split[1][i] = data.testCV(numberOfFolds, i);
    }

    return split;
}


    public static void main(String[] args) throws Exception 
    {
        BufferedReader datafile = readDataFile("training_1.arff");

        Instances data = new Instances(datafile);
        data.setClassIndex(data.numAttributes() - 1);

        // Do 10-split cross validation
        Instances[][] split = crossValidationSplit(data, 10);

        // Separate split into training and testing arrays
        Instances[] trainingSplits = split[0];
        Instances[] testingSplits = split[1];


        // Use a set of classifiers
        Classifier[] models = { 
//              new J48(), // a decision tree
//              new PART(), 
//                  new DecisionTable(),//decision table majority classifier
//          new DecisionStump(), //one-level decision tree
                new NaiveBayes(),
//                  new AdaBoostM1()
                new RandomForest()
//                  new LMT()
        };

        // Run for each model
        for (int j = 0; j < models.length; j++) 
        {

            // Collect every group of predictions for current model in a FastVector
            FastVector predictions = new FastVector();

            // For each training-testing split pair, train and test the classifier

            for (int i = 0; i < trainingSplits.length; i++) 
            {
                Evaluation validation = classify(models[j], trainingSplits[i], testingSplits[i]);

                predictions.appendElements(validation.predictions());
                System.out.println(validation.toMatrixString());
                // Uncomment to see the summary for each training-testing pair.
//              System.out.println(models[j].toString());
                 // generate curve
                ThresholdCurve tc = new ThresholdCurve();
                int classIndex = 0;
                Instances result = tc.getCurve(validation.predictions(), classIndex);
                System.out.println("tPR :"+validation.truePositiveRate(classIndex));
                System.out.println("fNR :"+validation.falseNegativeRate(classIndex));

                // plot curve
                ThresholdVisualizePanel vmc = new ThresholdVisualizePanel();
                vmc.setROCString("(Area under ROC = " + 
                    Utils.doubleToString(tc.getROCArea(result), 4) + ")");
                vmc.setName(result.relationName());
                PlotData2D tempd = new PlotData2D(result);
                tempd.setPlotName(result.relationName());
                tempd.addInstanceNumberAttribute();
                // specify which points are connected
                boolean[] cp = new boolean[result.numInstances()];
                for (int n = 1; n < cp.length; n++)
                  cp[n] = true;
                tempd.setConnectPoints(cp);
                // add plot
                vmc.addPlot(tempd);

                // display curve
                String plotName = vmc.getName(); 
                final javax.swing.JFrame jf = 
                  new javax.swing.JFrame("Weka Classifier Visualize: "+plotName);
                jf.setSize(500,400);
                jf.getContentPane().setLayout(new BorderLayout());
                jf.getContentPane().add(vmc, BorderLayout.CENTER);
                jf.addWindowListener(new java.awt.event.WindowAdapter() {
                  public void windowClosing(java.awt.event.WindowEvent e) {
                  jf.dispose();
                  }
                });
                jf.setVisible(true);

            }

            // Calculate overall accuracy of current classifier on all splits
            double accuracy = calculateAccuracy(predictions);

            // Print current classifier's name and accuracy in a complicated,
            // but nice-looking way.
            System.out.println("Accuracy of " + models[j].getClass().getSimpleName() + ": "
                    + String.format("%.2f%%", accuracy)
                    + "\n---------------------------------");


        }


    }
}

arff文件包含描述,后跟20个数据属性,后跟“是”&#39;或者&#39;否&#39;类标签。 在某些数据上运行此操作,TPR和FPR被准确地计算并显示为对应于每个混淆矩阵;但是,ROC曲线下面积显示为“NaN”。并且曲线是直的垂直或水平线: ROC Curve Image

我做错了什么?任何帮助将不胜感激。

1 个答案:

答案 0 :(得分:1)

这将是一个评论,但我是新的,无法发表评论。 我从循环内部运行你的代码,我的数据就像魅力一样。 所以这不是打印的问题。

看起来您的评估有效

  

在某些数据上运行此操作时,TPR和FPR会根据每个混淆矩阵进行精确计算和显示;

那你有没有尝试过评价课的以下功能?

evaluation.areaUnderROC(int classIndex);

要了解ROC曲线应该是什么样的?

您的标签类{“是”,“否”}或{0,1}? 我认为这不是问题,但你可以试试吗

Instances result = tc.getCurve(validation.predictions());
     

而不是

Instances result = tc.getCurve(validation.predictions(), classIndex);

你能发布一些混淆矩阵和TPR / FPR的值吗?

欢呼