Skip to content

[SPARK-5186] [MLLIB] Vector.equals and Vector.hashCode are very inefficient #3997

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
hhbyyh wants to merge 8 commits into from
Closed

[SPARK-5186] [MLLIB] Vector.equals and Vector.hashCode are very inefficient #3997

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
5741144
Specialized equals for SparseVector
Jan 13, 2015
f41b135
iterative equals for sparse vector
hhbyyh Jan 16, 2015
50abef3
fix ut for sparse vector with explicit 0
hhbyyh Jan 16, 2015
a6952c3
fix scala style for comments
hhbyyh Jan 16, 2015
bdf8789
improve equals and rewrite hashCode for Vector
hhbyyh Jan 17, 2015
985e160
improve locality for equals
hhbyyh Jan 17, 2015
93f0d46
unify sparse vs dense vectors
hhbyyh Jan 19, 2015
0d9d130
function name change and ut update
hhbyyh Jan 20, 2015
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
55 changes: 52 additions & 3 deletions mllib/src/main/scala/org/apache/spark/mllib/linalg/Vectors.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,13 +51,35 @@ sealed trait Vector extends Serializable {

override def equals(other: Any): Boolean = {
other match {
case v: Vector =>
util.Arrays.equals(this.toArray, v.toArray)
case v2: Vector => {
if (this.size != v2.size) return false
(this, v2) match {
case (s1: SparseVector, s2: SparseVector) =>
Vectors.equals(s1.indices, s1.values, s2.indices, s2.values)
case (s1: SparseVector, d1: DenseVector) =>
Vectors.equals(s1.indices, s1.values, 0 until d1.size, d1.values)
case (d1: DenseVector, s1: SparseVector) =>
Vectors.equals(0 until d1.size, d1.values, s1.indices, s1.values)
case (_, _) => util.Arrays.equals(this.toArray, v2.toArray)
Copy link
Member

@srowen srowen Jan 18, 2015

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yeah looking good. I have only minor things to say, which don't necessarily need to change. For example the utility method could be called Vectors.equals. Should there even be a catch-all case? I wonder if failing fast is good, if no other implementations are expected now.

Copy link
Contributor Author

@hhbyyh hhbyyh Jan 19, 2015

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for review, @srowen
Yes, I can change the name. Just like the name in util.Arrays.equals.
And for the catch-all case, I was thinking util.Arrays.equals is also fail-fast and it can cover the dense vs dense case well, and with some extensiblity. Maybe I don't catch your point. Let me know if still we should change this. Thanks

}
}
case _ => false
}
}

override def hashCode(): Int = util.Arrays.hashCode(this.toArray)
override def hashCode(): Int = {
var result: Int = size + 31
this.foreachActive { case (index, value) =>
// ignore explict 0 for comparison between sparse and dense
if (value != 0) {
result = 31 * result + index
// refer to {@link java.util.Arrays.equals} for hash algorithm
val bits = java.lang.Double.doubleToLongBits(value)
result = 31 * result + (bits ^ (bits >>> 32)).toInt
}
}
return result
}

/**
* Converts the instance to a breeze vector.
Expand Down Expand Up @@ -393,6 +415,33 @@ object Vectors {
}
squaredDistance
}

/**
* Check equality between sparse/dense vectors
*/
private[mllib] def equals(
v1Indices: IndexedSeq[Int],
v1Values: Array[Double],
v2Indices: IndexedSeq[Int],
v2Values: Array[Double]): Boolean = {
val v1Size = v1Values.size
val v2Size = v2Values.size
var k1 = 0
var k2 = 0
var allEqual = true
while (allEqual) {
while (k1 < v1Size && v1Values(k1) == 0) k1 += 1
while (k2 < v2Size && v2Values(k2) == 0) k2 += 1

if (k1 >= v1Size || k2 >= v2Size) {
return k1 >= v1Size && k2 >= v2Size // check end alignment
}
allEqual = v1Indices(k1) == v2Indices(k2) && v1Values(k1) == v2Values(k2)
k1 += 1
k2 += 1
}
allEqual
}
}

/**
Expand Down
18 changes: 18 additions & 0 deletions mllib/src/test/scala/org/apache/spark/mllib/linalg/VectorsSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,6 +89,24 @@ class VectorsSuite extends FunSuite {
}
}

test("vectors equals with explicit 0") {
val dv1 = Vectors.dense(Array(0, 0.9, 0, 0.8, 0))
val sv1 = Vectors.sparse(5, Array(1, 3), Array(0.9, 0.8))
val sv2 = Vectors.sparse(5, Array(0, 1, 2, 3, 4), Array(0, 0.9, 0, 0.8, 0))

val vectors = Seq(dv1, sv1, sv2)
for (v <- vectors; u <- vectors) {
assert(v === u)
assert(v.## === u.##)
}

val another = Vectors.sparse(5, Array(0, 1, 3), Array(0, 0.9, 0.2))
for (v <- vectors) {
assert(v != another)
assert(v.## != another.##)
}
}

Copy link
Member

@srowen srowen Jan 18, 2015

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This still needs tests involving dense vectors right?

Copy link
Contributor Author

@hhbyyh hhbyyh Jan 19, 2015

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sure. I'll add some.

test("indexing dense vectors") {
val vec = Vectors.dense(1.0, 2.0, 3.0, 4.0)
assert(vec(0) === 1.0)
Expand Down