#7 Error LabelPowerset with FilteredClassifier and StringToWordVector as a filter

[Anonymous]

Hi,

Mulan Version : 1.4.0

I am trying to use LabelPowerset with FilteredClassifier and StringToWordVector as a filter for text dataset, The same configuration is working fine with BinaryRelavance but with labelPowerset I am getting below exception msg :

Feb 23, 2015 12:29:56 PM mulan.classifier.transformation.LabelPowerset makePredictionInternal
SEVERE: null
java.lang.IndexOutOfBoundsException: Index: 26, Size: 26
at java.util.ArrayList.rangeCheck(Unknown Source)
at java.util.ArrayList.get(Unknown Source)
at weka.core.Attribute.value(Attribute.java:795)
at weka.core.AbstractInstance.stringValue(AbstractInstance.java:618)
at weka.core.AbstractInstance.stringValue(AbstractInstance.java:597)
at weka.filters.unsupervised.attribute.StringToWordVector.convertInstancewoDocNorm(StringToWordVector.java:1632)
at weka.filters.unsupervised.attribute.StringToWordVector.input(StringToWordVector.java:688)
at weka.classifiers.meta.FilteredClassifier.distributionForInstance(FilteredClassifier.java:424)
at mulan.classifier.transformation.LabelPowerset.makePredictionInternal(LabelPowerset.java:156)
at mulan.classifier.MultiLabelLearnerBase.makePrediction(MultiLabelLearnerBase.java:110)
at mulan.evaluation.Evaluator.evaluate(Evaluator.java:96)
at mulan.evaluation.Evaluator.evaluate(Evaluator.java:152)
at mulan.evaluation.Evaluator.innerCrossValidate(Evaluator.java:275)
at mulan.evaluation.Evaluator.crossValidate(Evaluator.java:238)
at mulan.wrapper.MulanCrossValidationWrapper.crossValidation(MulanCrossValidationWrapper.java:129)
at mulan.wrapper.MulanCrossValidationWrapper.evaluateAndGenerateOutputfromDir(MulanCrossValidationWrapper.java:58)
at mulan.execution.Execution.main(Execution.java:14)

Feb 23, 2015 12:29:56 PM mulan.evaluation.Evaluator innerCrossValidate
SEVERE: null
java.lang.NullPointerException
at mulan.core.Util.RandomIndexOfMax(Util.java:48)
at mulan.classifier.transformation.LabelPowerset.makePredictionInternal(LabelPowerset.java:161)
at mulan.classifier.MultiLabelLearnerBase.makePrediction(MultiLabelLearnerBase.java:110)
at mulan.evaluation.Evaluator.evaluate(Evaluator.java:96)
at mulan.evaluation.Evaluator.evaluate(Evaluator.java:152)
at mulan.evaluation.Evaluator.innerCrossValidate(Evaluator.java:275)
at mulan.evaluation.Evaluator.crossValidate(Evaluator.java:238)
at mulan.wrapper.MulanCrossValidationWrapper.crossValidation(MulanCrossValidationWrapper.java:129)
at mulan.wrapper.MulanCrossValidationWrapper.evaluateAndGenerateOutputfromDir(MulanCrossValidationWrapper.java:58)
at mulan.execution.Execution.main(Execution.java:14)

[tsoumakas]

Could not replicate this error with the latest version of Mulan and sample text data.

[Anonymous]

Hi,

The Mulan's sample text data (e.g. bibtex dataset) is already converted to StringToWordVector while I am using filtered classifier to do the task during classification.
I am attaching a sample dataset for you to run the test.
Kindly let me know if there is some issue with data format.
below is the code from main function :

public static void main(String[] args) throws Exception {
String arffFilename = Utils.getOption("arff", args);
String xmlFilename = Utils.getOption("xml", args);
MultiLabelInstances dataset = new MultiLabelInstances(arffFilename, xmlFilename);
FilteredClassifier classifier = new FilteredClassifier();
classifier.setClassifier(new NaiveBayesMultinomial());
classifier.setFilter(new StringToWordVectors());
LabelPowerset learner1 = new LabelPowerset(classifier);
Evaluator eval = new Evaluator();
MultipleEvaluation results;

    int numFolds = 4;
    results = eval.crossValidate(learner1, dataset, numFolds);
    System.out.println(results);
}

Thanks in advance :)

[Anonymous]

Hi,

The Sample dataset given on Mulan website is already converted to StringToWordVector (e.g bibtex dataset) while I am using FilteredClassifier to do the conversion internally during classification. I am attaching some sample data kindly use these files to run the experiment.
I am also pasting the sample main method used for cross validation :

public static void main(String[] args) throws Exception {
String arffFilename = Utils.getOption("arff", args);
String xmlFilename = Utils.getOption("xml", args);

    MultiLabelInstances dataset = new MultiLabelInstances(arffFilename, xmlFilename);
    FilteredClassifier classifier = new FilteredClassifier();
    classifier.setClassifier(new NaiveBayesMultinomial());
    classifier.setFilter(new StringToWordVectors());
    LabelPowerset learner1 = new LabelPowerset(classifier);
    Evaluator eval = new Evaluator();
    MultipleEvaluation results;

    int numFolds = 4;
    results = eval.crossValidate(learner1, dataset, numFolds);
    System.out.println(results);
}

[tsoumakas]

Hi,

Thanks for the sample data, I managed to replicate the bug.

It seems that this bug is due to Weka, as when I switched from weka-3.7.10.jar to weka-3.7.12.jar it dissapeared.

Best,
Greg

[Anonymous]

Hi,

Thanks a lot.

Please let me know which version of Mulan you are using.. I am currently using Mulan-1.4.0 which I think not compatible with Weka-3.7.12

Thanks :)

[tsoumakas]

Hi,

I'm using the latest Mulan-1.5.0, but I believe that Mulan-1.4.0 is also compatible with weka-3.7.12

Greg

[apoorv]

Hi Greg,

I tried multiple times with the updated versions as you mentioned but I am still getting the same error.

can you please check once again for cross validation and confirm the error is gone.

Thanks

[tsoumakas]

Hi,

Apologies for the misunderstanding. Upgrading to Weka 3.7.12 fixed the problem for simple holdout evaluation but not for cross-validation

There was a bug in the LabelPowerset algorithm after all, which has now been fixed. You should grab the latest code from our Git repository.

Greg