如何使用带有 Java 代码的 WEKA 评估类来显示标准差值

Question

我正在编写一个 WEKA 分类器。我使用了评估类 (http://weka.sourceforge.net/doc.dev/weka/classifiers/Evaluation.html) 的 toSummaryString()、toMatrixString 和 toClassDetailString 方法来显示结果。结果显示正确。但是，我想显示每个结果的标准偏差值，因为它们是通过使用 WEKA 的实验者 GUI 显示的。我怎样才能做到这一点 ？

Answer 1

要获得标准差，您需要使用Experiment 类，然后获得统计信息。这是一个示例：Using the Experiment API。

Answer 2

您可以使用以下代码：

package weka.api;
//import required classes
import weka.experiment.Stats;
import weka.core.AttributeStats;
import weka.core.Instance;
import weka.core.Instances;
import weka.core.converters.ConverterUtils.DataSource;

public class AttInst {
    public static void main(String args[]) throws Exception{
        //load dataset
        DataSource source = new DataSource("D:/y.arff");
        //get instances object 
        Instances data = source.getDataSet();
        //set class index .. as the last attribute
        if (data.classIndex() == -1) {
           data.setClassIndex(data.numAttributes() - 1);
        }
        //get number of attributes (notice class is not counted)
        int numAttr = data.numAttributes() - 1;
        for (int i = 0; i < numAttr; i++) {
            //check if current attr is of type nominal
            if (data.attribute(i).isNominal()) {
                System.out.println("The "+i+"th Attribute is Nominal"); 
                //get number of values
                int n = data.attribute(i).numValues();
                System.out.println("The "+i+"th Attribute has: "+n+" values");
            }           

            //get an AttributeStats object
            AttributeStats as = data.attributeStats(i);
            int dC = as.distinctCount;
            System.out.println("The "+i+"th Attribute has: "+dC+" distinct values");

            //get a Stats object from the AttributeStats
            if (data.attribute(i).isNumeric()){
                System.out.println("The "+i+"th Attribute is Numeric"); 
                Stats s = as.numericStats;
                System.out.println("The "+i+"th Attribute has min value: "+s.min+" and max value: "+s.max+" and mean value: "+s.mean+" and stdDev value: "+s.stdDev );
            }

    }


    }
}