trainActual()

TrainNewsGroupsKMeansLogisticRegression.java

  public static void trainActual(Iterable<Pair<Text, VectorWritable>> inputIterable, String outBase,
                                 Map<String, Integer> clusterNamesToIds) throws  IOException {
    Map<String, Centroid> actualClusters = Maps.newHashMap();
    computeActualClusters(inputIterable, actualClusters);

    OnlineLogisticRegression learningAlgorithm =
        new OnlineLogisticRegression(NUM_CLASSES, NUM_FEATURES_ACTUAL, new L1());

    for (Pair<Text, VectorWritable> pair : inputIterable) {
      Vector actualCentroid = pair.getSecond().get();
      Vector features = new DenseVector(NUM_FEATURES_ACTUAL);
      int i = 0;
      for (Centroid centroid : actualClusters.values()) {
        features.set(i++, centroid.getDistanceSquared(actualCentroid));
      }
      String clusterName = pair.getFirst().toString();
      System.out.printf("Feature vector of size %d; NUM_FEATURES_ACTUAL: %d; %d\n",
          features.size(), NUM_FEATURES_ACTUAL, clusterNamesToIds.get(clusterName));
      learningAlgorithm.train(clusterNamesToIds.get(clusterName), features);
    }
    learningAlgorithm.close();

    ModelSerializer.writeBinary(outBase + "-append.model", learningAlgorithm);
  }