spark-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From jkbradley <...@git.apache.org>
Subject [GitHub] spark pull request: [SPARK-4797] Replace breezeSquaredDistance
Date Tue, 16 Dec 2014 20:32:35 GMT
Github user jkbradley commented on a diff in the pull request:

    https://github.com/apache/spark/pull/3643#discussion_r21929723
  
    --- Diff: mllib/src/main/scala/org/apache/spark/mllib/util/MLUtils.scala ---
    @@ -264,6 +263,86 @@ object MLUtils {
         }
         Vectors.fromBreeze(vector1)
       }
    + 
    +  /**
    +   * Returns the squared distance between two Vectors.
    +   */
    +  private[util] def vectorSquaredDistance(v1: Vector, v2: Vector): Double = {
    +    var squaredDistance = 0.0
    +    (v1, v2) match { 
    +      case (v1: SparseVector, v2: SparseVector) =>
    +        val v1Values = v1.values
    +        val v1Indices = v1.indices
    +        val v2Values = v2.values
    +        val v2Indices = v2.indices
    +        val nnzv1 = v1Indices.size
    +        val nnzv2 = v2Indices.size
    +        
    +        var kv1 = 0
    +        var kv2 = 0
    +        while (kv1 < nnzv1 || kv2 < nnzv2) {
    +          var score = 0.0
    + 
    +          if (kv2 >= nnzv2 || (kv1 < nnzv1 && v1Indices(kv1) < v2Indices(kv2)))
{
    +            score = v1Values(kv1)
    +            kv1 += 1
    +          } else if (kv1 >= nnzv1 || (kv2 < nnzv2 && v2Indices(kv2) <
v1Indices(kv1))) {
    +            score = v2Values(kv2)
    +            kv2 += 1
    +          } else if ((kv1 < nnzv1 && kv2 < nnzv2) && v1Indices(kv1)
== v2Indices(kv2)) {
    +            score = v1Values(kv1) - v2Values(kv2)
    +            kv1 += 1
    +            kv2 += 1
    +          }
    +          squaredDistance += score * score
    +        }
    +
    +      // The following two cases are used to handle dense and approximately dense vectors
    +      case (v1: SparseVector, v2: DenseVector) if v1.indices.length / v1.size < 0.5
=>
    +        squaredDistance = vectorSquaredDistance(v1, v2)
    +
    +      case (v1: DenseVector, v2: SparseVector) if v2.indices.length / v2.size < 0.5
=>
    +        squaredDistance = vectorSquaredDistance(v2, v1)
    +
    +      case (v1, v2) =>
    +        squaredDistance = v1.toArray.zip(v2.toArray).foldLeft(0.0){(distance, elems)
=>
    +          val score = elems._1 - elems._2
    +          distance + score * score
    +        }
    +    }
    +    squaredDistance
    +  }
    +
    +  /**
    +   * Returns the squared distance between DenseVector and SparseVector.
    +   */
    +  private[util] def vectorSquaredDistance(v1: SparseVector, v2: DenseVector): Double
= {
    +    var kv1 = 0
    +    var kv2 = 0
    +    var indices = v1.indices
    +    var squaredDistance = 0.0
    +    var iv1 = indices(kv1)
    +    val nnzv2 = v2.size
    +
    +    while (kv2 < nnzv2) {
    +      var score = 0.0
    +      if (kv2 < iv1 || kv2 > iv1) {
    +        score = v2(kv2)
    +        squaredDistance += score * score
    +      }
    +      if (kv2 == iv1 && kv1 < indices.length) {
    --- End diff --
    
    Use ```else if``` for efficiency (and so you can skip testing ```kv2 == iv1```).


---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org


Mime
View raw message