Mahout-从Cluster看类扩展理念

最顶层的接口是最抽象的Model抽象，代码中给出了定义，模型是数据的概率分布

模型具有的功能是返回某个观察值的概率，（带权）吸收观察值，计算模型参数，统计观察值数量，返回采样

public interface Model<O> extends Writable {
  double pdf(O x);
  void observe(O x);
  void observe(O x, double weight);
  void computeParameters();
  long count();
  Model<VectorWritable> sampleFromPosterior();
}

聚类Cluster是模型的一种，可以返回id、中心、半径、数量、生成易读字符串描述

public interface Cluster extends Model<VectorWritable>, Parametered {
  int getId();
  Vector getCenter();
  Vector getRadius();
  long getNumPoints();
  String asFormatString(String[] bindings);
}

AbstractCluster抽象类实现接口，所有的构造函数是protected，意味着不能直接实例化，只能实例化子类，类名称Abstract反映出了此种特性

public abstract class AbstractCluster implements Cluster

由于本质还是Writable的，实现了读写方法

  @Override
  public void readFields(DataInput in) throws IOException {
    this.id = in.readInt();
    this.numPoints = in.readLong();
    VectorWritable temp = new VectorWritable();
    temp.readFields(in);
    this.center = temp.get();
    temp.readFields(in);
    this.radius = temp.get();
  }
  
  @Override
  public void write(DataOutput out) throws IOException {
    out.writeInt(id);
    out.writeLong(numPoints);
    VectorWritable.writeVector(out, center);
    VectorWritable.writeVector(out, radius);
  }

java中抽象类可以不必实现接口的所有方法，有一些方法没有实现，比如pdf方法

 

DistanceMeasureCluster实现类，扩展了度量方法，聚类的度量方法是多样的，用ClassLoader加载度量类类

public class DistanceMeasureCluster extends AbstractCluster

非抽象类必须实现接口全部函数，因此实现了先前没实现过的pdf函数
在实现读写函数时，用了super方法，在原基础上追加内容

  @Override
  public void readFields(DataInput in) throws IOException {
    String dm = in.readUTF();
    try {
      ClassLoader ccl = Thread.currentThread().getContextClassLoader();
      this.measure = ccl.loadClass(dm).asSubclass(DistanceMeasure.class).newInstance();
    } catch (InstantiationException e) {
      throw new IllegalStateException(e);
    } catch (IllegalAccessException e) {
      throw new IllegalStateException(e);
    } catch (ClassNotFoundException e) {
      throw new IllegalStateException(e);
    }
    super.readFields(in);
  }

  @Override
  public void write(DataOutput out) throws IOException {
    out.writeUTF(measure.getClass().getName());
    super.write(out);
  }

 具体到kmeans聚类下，又扩展了判断收敛的函数，相应的读写也用super方法追加新内容

public class Cluster extends DistanceMeasureCluster

  @Override
  public void write(DataOutput out) throws IOException {
    super.write(out);
    out.writeBoolean(converged);
  }

  @Override
  public void readFields(DataInput in) throws IOException {
    super.readFields(in);
    this.converged = in.readBoolean();
  }

以上可以看出从抽象到具体，从基础功能逐步扩展新功能的一套完整的体系~