=klassy=

Weka classification wrapper

Description

• Klassy provides a Weka friendly API with only two methods
• Intended for using existing Weka models in production code
• Build against Weka 3.8.1 (but can be used with earlier versions or the stripped Android version)

Installation

repositories {
    jcenter()
}
...
dependencies {    
    compile('com.maslick.ai:klassy:0.1.8')
}

Usage (see example)

• extend the AbstractClassifier class and implement two methods - createAttributeList and calculateFeatures
• depending on your deployment target (pc, android, etc.) implement the IFileLoader interface (see below)
• instantiate your model (classifier) and invoke its classify method:

@Test
public void housing() {
    Houser model = new Houser(new ContextLoader());
    House house = House.builder()
                    .houseSize(3198)
                    .lotSize(9669)
                    .bedrooms(5)
                    .granite(2)
                    .bathroom(1)
                    .build();
    String klass = model.classify(house);
    Assert.assertEquals(219328, Double.valueOf(klass), 1);
}

public class Houser extends AbstractClassifier<House> {

    public Houser(IFileLoader fileLoader) {
        super(fileLoader);
        MODEL = "house.model";         // model file name without path
        classifierType = REGRESSION;   // CLASSIFICATION or REGRESSION
        classIndex = 5;                // usually comes last (5) or first (0)
        relation = "house";            // name of the dataset (optional)
    }

    @Override
    public ArrayList<Attribute> createAttributeList() {
        ArrayList<Attribute> atts = new ArrayList<>();

        // Weka doesn't take attribute names into account, but their order!
        // However, for clarity one should specify attribute names like below

        atts.add(new Attribute("houseSize",    0));
        atts.add(new Attribute("lotSize",      1));
        atts.add(new Attribute("bedrooms",     2));
        atts.add(new Attribute("granite",      3));
        atts.add(new Attribute("bathroom",     4));
        atts.add(new Attribute("sellingPrice", 5));    // class attribute (classIndex=5)

        return atts;
    }

    @Override
    public Instance calculateFeatures(House data) {
        // valid instance size is 5 or 6 (with class attribute)
        Instance instance = new DenseInstance(5);

        instance.setValue(attributes.get(0), data.getHouseSize());
        instance.setValue(attributes.get(1), data.getLotSize());
        instance.setValue(attributes.get(2), data.getBedrooms());
        instance.setValue(attributes.get(3), data.getGranite());
        instance.setValue(attributes.get(4), data.getBathroom());
        
        return instance;
    }
}

---

IFileLoader

IFileLoader interface allows you to abstract away from your deployment target and load weka models (it may well be a PC, as well as the mobile phone or cloud).

public interface IFileLoader {
    InputStream getFile(String filename) throws IOException;
}

SpringBoot implementation:

public class SpringFileLoader implements IFileLoader {
    @Override
    public InputStream getFile(String filename) throws IOException {
        return new ClassPathResource(filename).getInputStream();
    }
}

Android implementation:

public class AndroidFileLoader implements IFileLoader {
    private Context context;
    
    public AndroidFileLoader(Context context) {
        this.context = context;
    }

    @Override
    public InputStream getFile(String filename) throws IOException {
        context.getApplicationContext().getAssets().open(filename);
    }
}

Klassy provides the default ContextLoader implementation:

public class ContextLoader implements IFileLoader{
    @Override
    public InputStream getFile(String filename) throws IOException {
        return ClassLoader.getSystemResourceAsStream(filename);
    }
}