Weka开发[4]-特征选择

 特征选择,我对这一部分也不熟,大概讲一下,用AttributeSelection进行特征选择,它需要设置3个方面,第一:对属性评价的类(自己到Weka软件里看一下,英文Attribute Evaluator),第二:搜索的方式(自己到Weka软件里看一下,英文Search Method),第三:就是你要进行特征选择的数据集了。最后调用Filter的静态方法userFilter,感觉写的都是废话,一看代码就明白了。唯一值得一说的也就是别把AttributeSelection的包加错了,代码旁边有注释。

       另一个函数懒的解释了(它也不是我写的),基本上是自解释的,不太可能看不懂。

 

package instanceTest;

 

import java.io.FileReader;

import java.util.Random;

 

import weka.attributeSelection.CfsSubsetEval;

import weka.attributeSelection.GreedyStepwise;

import weka.classifiers.Evaluation;

import weka.classifiers.meta.AttributeSelectedClassifier;

import weka.classifiers.trees.J48;

import weka.core.Instances;

import weka.filters.Filter;

import weka.filters.supervised.attribute.AttributeSelection;

 

public class FilterTest

{

    private Instances m_instances = null;

   

    public void getFileInstances( String fileName ) throws Exception

    {

        FileReader frData = new FileReader( fileName );

        m_instances = new Instances( frData );

       

        m_instances.setClassIndex( m_instances.numAttributes() - 1 );

    }

   

    public void selectAttUseFilter() throws Exception

    {

        AttributeSelection filter = new AttributeSelection();  // package weka.filters.supervised.attribute!

        CfsSubsetEval eval = new CfsSubsetEval();

        GreedyStepwise search = new GreedyStepwise();

        filter.setEvaluator(eval);

        filter.setSearch(search);

        filter.setInputFormat( m_instances );

       

        System.out.println( "number of instance attribute = " +m_instances.numAttributes() );

       

        Instances selectedIns = Filter.useFilterm_instances, filter);

        System.out.println( "number of selected instance attribute = " + selectedIns.numAttributes() );

    }

   

    public void selectAttUseMC() throws Exception

    {  

         AttributeSelectedClassifier classifier = newAttributeSelectedClassifier();

         CfsSubsetEval eval = new CfsSubsetEval();

         GreedyStepwise search = new GreedyStepwise();

         J48 base = new J48();

         classifier.setClassifier( base );

         classifier.setEvaluator( eval );

         classifier.setSearch( search );

         // 10-fold cross-validation

         Evaluation evaluation = new Evaluation( m_instances );

         evaluation.crossValidateModel(classifier, m_instances, 10, newRandom(1));

         System.out.println( evaluation.toSummaryString() );

    }

   

    public static void main( String[] args ) throws Exception

    {

        FilterTest filter = new FilterTest();

       

        filter.getFileInstances( "F://Program Files//Weka-3-4//data//soybean.arff");

        filter.selectAttUseFilter();

       

        filter.selectAttUseMC();

    }

 

}

你可能感兴趣的:(开发)