Distance Metric Implementation Example

gistfile1.scala

import breeze.generic.UFunc
import breeze.linalg.{sum, DenseVector => DBV, Vector => BV}
import breeze.macros.expand

object distance extends UFunc {

  val DEFAULT_METHOD = "euclidean"

  @expand
  @expand.valify
  implicit def distanceImpl: Impl3[BV[Double], BV[Double], String, Double] = {
    new Impl3[BV[Double], BV[Double], String, Double] {
      def apply(v1: BV[Double], v2: BV[Double], method: String): Double = {
        method match {
          case "euclidean" => euclidean(v1, v2)
          case "manhattan" => manhattan(v1, v2)
          case _ => euclidean(v1, v2) // originally throws
        }
      }
    }
  }

  @expand
  @expand.valify
  implicit def defaultDistanceImpl: Impl2[BV[Double], BV[Double], Double] = {
    new Impl2[BV[Double], BV[Double], Double] {
      def apply(v1: BV[Double], v2: BV[Double]): Double = {
        distance(v1, v2, DEFAULT_METHOD)
      }
    }
  }
}

object euclidean extends UFunc {
  @expand
  @expand.valify
  implicit def distanceImpl: Impl2[BV[Double], BV[Double], Double] = {
    new Impl2[BV[Double], BV[Double], Double] {
      def apply(v1: BV[Double], v2: BV[Double]): Double = {
        val d = v1 - v2
        Math.sqrt(d dot d)
      }
    }
  }
}

object manhattan extends UFunc {
  @expand
  @expand.valify
  implicit def distanceImpl: Impl2[BV[Double], BV[Double], Double] = {
    new Impl2[BV[Double], BV[Double], Double] {
      def apply(v1: BV[Double], v2: BV[Double]): Double = {
        sum((v1 - v2).map(Math.abs))
      }
    }
  }
}

Thanks!

1. no reason to use @expand if you're not using type parameters (@expand code gens type parameters)
2. manhattan can just be norm(v1 - v2, 1), euclidean norm(v1 - v2, 2)
3. I'd rather use objects in a sealed (or even non-sealed) trait than strings

HI @dlwh,

Thank you for your comment.
I am working to contribute some distance functions to Breeze.
I will send a PR to Breeze, please wait a moment.