weka中的自定义批量过滤器

Question

我正在尝试构建一个扩展 SimpleBatchFilter 的自定义批处理过滤器。但是，我遇到了第二次运行它以获得反转输出的问题。这是相关代码和两次运行完成后我得到的错误：

Exception in thread "main" java.lang.IndexOutOfBoundsException: Index: 79, Size: 79
    at java.util.ArrayList.rangeCheck(ArrayList.java:653)
    at java.util.ArrayList.get(ArrayList.java:429)
    at weka.core.Attribute.addStringValue(Attribute.java:994)
    at weka.core.StringLocator.copyStringValues(StringLocator.java:155)
    at weka.core.StringLocator.copyStringValues(StringLocator.java:91)
    at weka.filters.Filter.copyValues(Filter.java:373)
    at weka.filters.Filter.push(Filter.java:290)
    at weka.filters.SimpleBatchFilter.batchFinished(SimpleBatchFilter.java:266)
    at weka.filters.Filter.useFilter(Filter.java:667)
    at likeability.Main.main(Main.java:30)

这里是相关代码：

public class TestFilter extends SimpleBatchFilter {

    private Attribute a;
    private Attribute b;
    private int sampleSizePercent = 15;
    private boolean invert = false;
    private int seed = 1;

    @Override
    protected Instances process(Instances inst) throws Exception {
        ArrayList<Instances> partitionsA = partition(inst, a);
        ArrayList<Instances> partitions = new ArrayList<Instances>();
        for(Instances data: partitionsA) {
            partitions.addAll(partition(data, b));
        }

        return getTestSet(partitions);
    }

    /*
     * Partitions the data so that there's only one nominal value of the
     * attribute a in one partition.
     */
    private ArrayList<Instances> partition(Instances data, Attribute att) throws Exception {
        ArrayList<Instances> instances = new ArrayList<Instances>();
        for (int i = 0; i < att.numValues(); i++){
            RemoveWithValues rm = new RemoveWithValues();
            rm.setAttributeIndex(Integer.toString(att.index()+1));
            rm.setInvertSelection(true);
            rm.setNominalIndices(Integer.toString(i+1));
            rm.setInputFormat(data);
            instances.add(Filter.useFilter(data, rm));
        }
        return instances;
    }

    private Instances getTestSet(List<Instances> insts) throws Exception {
        Instances output = new Instances(insts.get(0), 0);

        for(Instances inst: insts) {
            Resample filter = new Resample();
            filter.setRandomSeed(seed);
            filter.setNoReplacement(true);
            filter.setInvertSelection(invert);
            filter.setSampleSizePercent(sampleSizePercent);
            filter.setInputFormat(inst);
            Instances curr = Filter.useFilter(inst, filter);
            System.out.println(inst.size() + " " + curr.size());
            output.addAll(curr);
        }
        return output;

    }

    @Override
    protected Instances determineOutputFormat(Instances arg) throws Exception {
        return new Instances(arg, 0);
    }

    @Override
    public String globalInfo() {
        return "A filter which partitions the data so that each partition contains"
                + " only instances with one value of attribute a and b, then takes "
                + "a random subset of values from each partition and merges them to"
                + " produce the final set.";
    }

    public Capabilities getCapabilities() {
         Capabilities result = super.getCapabilities();
         result.enableAllAttributes();
         result.enableAllClasses();
         result.enable(Capability.NO_CLASS);  // filter doesn't need class to be set
         return result;
       }
   //Main and getters and setters

}

这就是我所说的：

    TestFilter filter = new TestFilter();
    filter.setA(data.attribute("gender"));
    filter.setB(data.attribute("age"));
    filter.setInputFormat(data);
    Instances test = Filter.useFilter(data, filter);
    filter.setInvert(true);
    filter.setInputFormat(data);
    Instances train = Filter.useFilter(data, filter);

在我看来，我需要在调用之间使用这两行代码是很愚蠢的。我怀疑我应该使用 isBatchFinished()，这是否意味着我必须扩展 BatchFilter 而不是 SimpleBatchFilter 来实现它？看到一些成功的实现也会很有帮助，因为我能在 WEKA 手册中找到唯一的实现。

Answer 1

我通过扩展过滤器并将处理函数更改为 batchFinished() 来解决它。我发布此答案是因为我在其他任何地方都没有找到自定义过滤器示例。

@Override
public boolean batchFinished() throws Exception {
    if(isFirstBatchDone()) {
        invert = true;
    }
    if (getInputFormat() == null)
        throw new NullPointerException("No input instance format defined");
    Instances inst = getInputFormat();
    ArrayList<Instances> partitionsA = partition(inst, a);
    ArrayList<Instances> partitions = new ArrayList<Instances>();
    for(Instances data: partitionsA) {
        partitions.addAll(partition(data, b));
    }

private void getTestSet(List<Instances> insts) throws Exception {

    for(Instances inst: insts) {
        Resample filter = new Resample();
        filter.setRandomSeed(seed);
        filter.setNoReplacement(true);
        filter.setInvertSelection(invert);
        filter.setSampleSizePercent(sampleSizePercent);
        filter.setInputFormat(inst);
        Instances curr = Filter.useFilter(inst, filter);
        System.out.println(inst.size() + " " + curr.size());
        curr.forEach((i) -> push(i));
    }
}

@Override
public boolean setInputFormat(Instances arg) throws Exception {
    super.setInputFormat(arg);
    Instances outputFormat = new Instances(arg, 0);
    setOutputFormat(outputFormat);
    return true;
}