zlibsvm is an object-oriented, easy to use and simple Java binding for the famous LIBSVM library hosted on GitHub.
It encapsulates the cross-compiled Java code from LIBSVM behind an object-oriented API which can be easily used via Apache Maven in your own projects.
To use the latest release of zlibsvm, please use the following snippet in your pom.xml
<dependency>
<groupId>de.hs-heilbronn.mi</groupId>
<artifactId>zlibsvm-core</artifactId>
<version>2.1.0</version>
</dependency>A code example can be found here.
The dataset format for LIBSVM is
label feature_id1:feature_value1 feature_id2:feature_value2 ...
Thus, every feature (or value) needs its own unique identifier.
For three different class labels 1,2,3 and a feature set consisting of a(id=1),b(id=2),c=(id=3), a valid data representation for three data points d1,d2,d3 would be
2 1:0.5325 3:0.523
3 2:0.7853 3:0.6326
1 1:0.53265 2:0.5422
Meaning:
d1contains featurea(id=1)andc(id=3)d2contains featureb(id=2)andc(id=3)d3contains featurea(id=1)andb(id=2)
Note, that it is not necessary to provide feature_id1:feature_value1 for features, which are not contained in the given data point.
First of all, you need to implement your custom SvmDocument and a custom SvmFeature, which could be like:
public class SvmDocumentImpl implements SvmDocument {
private final List<SvmFeature> features;
private final List<SvmClassLabel> classLabels = new ArrayList<>();
public SvmDocumentImpl(List<SvmFeature> features) {
this.features = features;
}
public List<SvmFeature> getSvmFeatures() {
return features;
}
public SvmClassLabel getClassLabelWithHighestProbability() {
if (classLabels.isEmpty()) {
return null;
}
return Collections.max(classLabels);
}
@Override
public List<SvmClassLabel> getAllClassLabels() {
return Collections.unmodifiableList(classLabels);
}
@Override
public void addClassLabel(SvmClassLabel classLabel) {
assert (classLabel != null);
this.classLabels.add(classLabel);
}
}
public record SvmFeatureImpl(int index, double value) implements SvmFeature {
public int getIndex() {
return index;
}
public double getValue() {
return value;
}
@Override
public int compareTo(SvmFeature o) {
return Integer.compare(getIndex(), o.getIndex());
}
}To obtain an SvmModel the SVM needs to be trained. This is done via an SvmConfigurationImpl.Builder(), which is used to specify your custom SVM configuration.
The default configuration is the same as described here: C_SVC, RBF-Kernel with gamma 0 and cost 1.
SvmTrainer trainer = new SvmTrainerImpl(new SvmConfigurationImpl.Builder().build(),"my-custom-trained-model");
SvmModel model = trainer.train(documentsForTraining);After this step it is possible to use this SvmModel for prediction.
SvmClassifier classifier = new SvmClassifierImpl(model);
List<SvmDocument> classified = classifier.classify(documentsForPrediction, true);
for(SvmDocument d : classified) {
System.out.println(d.toString() " was classified as category:" d.getClassLabelWithHighestProbability().getNumeric());
}
Published under Apache License 2.0