0

此类（https://github.com/scalanlp/nak/blob/20637cfd6f373792b7d2fb4d9c965c0a6dac89d9/src/main/scala/nak/cluster/Kmeans.scala）作为其构造函数：

class Kmeans(
  points: IndexedSeq[Point],
  distance: DistanceFunction,
  minChangeInDispersion: Double = 0.0001,
  maxIterations: Int = 100,
  fixedSeedForRandom: Boolean = false
)

距离函数是一个具有特征的对象（https://github.com/scalanlp/nak/blob/ae8fc0c534ea0613300e8c53487afe099327977a/src/main/scala/nak/cluster/Points.scala）：

trait DistanceFunction extends ((Point, Point) => Double)

/**
 * A companion object to the DistanceFunction trait that helps select the
 * DistanceFunction corresponding to each string description.
 */
object DistanceFunction {
  def apply(description: String) = description match {
    case "c" | "cosine" => CosineDistance
    case "m" | "manhattan" => ManhattanDistance
    case "e" | "euclidean" => EuclideanDistance
    case _ => throw new MatchError("Invalid distance function: " + description)
  }
}

/**
 * Compute the cosine distance between two points. Note that it is a distance
 * because we subtract the cosine similarity from one.
 */
object CosineDistance extends DistanceFunction {
  def apply(x: Point, y: Point) = 1 - x.dotProduct(y) / (x.norm * y.norm)
}

/**
 * Compute the Manhattan (city-block) distance between two points.
 */
object ManhattanDistance extends DistanceFunction {
  def apply(x: Point, y: Point) = (x - y).abs.sum
}

/**
 * Compute the Euclidean distance between two points.
 */
object EuclideanDistance extends DistanceFunction {
  def apply(x: Point, y: Point) = (x - y).norm
}

到目前为止，这是我的构造函数实现：

val p1 = new Point(IndexedSeq(0.0, 0.0 , 3.0));
val p2 = new Point(IndexedSeq(0.0, 0.0 , 3.0));
val p3 = new Point(IndexedSeq(0.0, 0.0 , 3.0));

val clusters1 =  IndexedSeq( p1 , p2 , p3 )

val k = new Kmeans(clusters1 , ??????

如何创建 DistanceFunction 实现以实现 Kmeans 构造函数？我可以只使用现有的对象 DistanceFunction 吗？

4

1 回答 1

1

您不能使用 DistanceFunction 伴生对象——它只是放置相应特征/类的静态变量的地方，但您可以使用提供的扩展特征的对象：CosineDistance、ManhattanDistance 或 EuclideanDistance：

val k = new Kmeans(clusters1 , CosineDistance, ... )

如果你想创建自己的实现，你已经有一个例子：

object ManhattanDistance extends DistanceFunction {
  def apply(x: Point, y: Point) = (x - y).abs.sum
}

或与类：

class ManhattanDistance extends DistanceFunction {
  def apply(x: Point, y: Point) = (x - y).abs.sum
}
val k = new Kmeans(clusters1 , new ManhattanDistance(), ... )
于 2013-07-12T14:44:22.773 回答