OpenNLP MaxEnt - ContextGenerator and EventStream

Question

I want to write my own model using openNLP MaxEnt, for that I want to implement ContextGenerator and EventStream interfaces(as mentioned in documentation). I looked at the these implementations for openNLP Chuncker, POSTagger and NameFinder, but all those implementations used 'Pair' which is deprecated and by just looking at the code I don't understand what their respective ContextGenerators are doing. The model that I will be creating will classify each token as a RoomNumber or not a RoomNumber by looking at POS tags for each token. How should I start coding ContextGenerator and EventStream for this model. I know what a context is and what a feature is, but I don't know what a ContextGenerator do and what an EvenStream do. I did look at openNLP maxent page, but it is not helpful. Please help me understand this, thank you.

Answer 1

The following code might help, although it does not use the ContextGenerator explicitly. Actually, the BasicContextGenerator is used within the BasicEventStream and it just splits each input string into a list of features.

e.g. the String "a=1 b=2 c=1" is split into 3 features: "a=1", "b=2" and "c=1".

If you just want to use the Maxent API to train the model and then to use it for classification, you can use the following approach that worked for me:

package opennlptest;

import java.io.IOException;
import java.util.Arrays;
import java.util.List;

import opennlp.maxent.GIS;
import opennlp.model.Event;
import opennlp.model.EventStream;
import opennlp.model.ListEventStream;
import opennlp.model.MaxentModel;

public class TestMaxentEvents {

    static Event createEvent(String outcome, String... context) {
        return new Event(outcome, context);
    }

    public static void main(String[] args) throws IOException {

        // here are the input training samples
        List<Event> samples =  Arrays.asList(new Event[] {
                //           outcome + context
                createEvent("c=1", "a=1", "b=1"),
                createEvent("c=1", "a=1", "b=0"),
                createEvent("c=0", "a=0", "b=1"),
                createEvent("c=0", "a=0", "b=0")
        });

        // training the model
        EventStream stream = new ListEventStream(samples);
        MaxentModel model = GIS.trainModel(stream);

        // using the trained model to predict the outcome given the context
        String[] context = {"a=1", "b=0"};
        double[] outcomeProbs = model.eval(context);
        String outcome = model.getBestOutcome(outcomeProbs);

        System.out.println(outcome); // output: c=1
    }

}