From 5994cfe81271a39294aa29fd47aa94c99aa56743 Mon Sep 17 00:00:00 2001
From: Nick Pritchard <nicholas.pritchard@falkonry.com>
Date: Thu, 8 Oct 2015 22:22:20 -0700
Subject: [PATCH] [SPARK-10875] [MLLIB] Computed covariance matrix should be
symmetric
Compute upper triangular values of the covariance matrix, then copy to lower triangular values.
Author: Nick Pritchard <nicholas.pritchard@falkonry.com>
Closes #8940 from pnpritchard/SPARK-10875.
---
.../mllib/linalg/distributed/RowMatrix.scala | 6 ++++--
.../linalg/distributed/RowMatrixSuite.scala | 18 ++++++++++++++++++
2 files changed, 22 insertions(+), 2 deletions(-)
diff --git a/mllib/src/main/scala/org/apache/spark/mllib/linalg/distributed/RowMatrix.scala b/mllib/src/main/scala/org/apache/spark/mllib/linalg/distributed/RowMatrix.scala
index 7c7d900af3..b8a7adceb1 100644
--- a/mllib/src/main/scala/org/apache/spark/mllib/linalg/distributed/RowMatrix.scala
+++ b/mllib/src/main/scala/org/apache/spark/mllib/linalg/distributed/RowMatrix.scala
@@ -357,9 +357,11 @@ class RowMatrix @Since("1.0.0") (
var alpha = 0.0
while (i < n) {
alpha = m / m1 * mean(i)
- j = 0
+ j = i
while (j < n) {
- G(i, j) = G(i, j) / m1 - alpha * mean(j)
+ val Gij = G(i, j) / m1 - alpha * mean(j)
+ G(i, j) = Gij
+ G(j, i) = Gij
j += 1
}
i += 1
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/linalg/distributed/RowMatrixSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/linalg/distributed/RowMatrixSuite.scala
index 283ffec1d4..4abb98fb6f 100644
--- a/mllib/src/test/scala/org/apache/spark/mllib/linalg/distributed/RowMatrixSuite.scala
+++ b/mllib/src/test/scala/org/apache/spark/mllib/linalg/distributed/RowMatrixSuite.scala
@@ -24,6 +24,7 @@ import breeze.linalg.{DenseVector => BDV, DenseMatrix => BDM, norm => brzNorm, s
import org.apache.spark.SparkFunSuite
import org.apache.spark.mllib.linalg.{Matrices, Vectors, Vector}
+import org.apache.spark.mllib.random.RandomRDDs
import org.apache.spark.mllib.util.{LocalClusterSparkContext, MLlibTestSparkContext}
class RowMatrixSuite extends SparkFunSuite with MLlibTestSparkContext {
@@ -255,6 +256,23 @@ class RowMatrixSuite extends SparkFunSuite with MLlibTestSparkContext {
assert(closeToZero(abs(expected.r) - abs(rOnly.R.toBreeze.asInstanceOf[BDM[Double]])))
}
}
+
+ test("compute covariance") {
+ for (mat <- Seq(denseMat, sparseMat)) {
+ val result = mat.computeCovariance()
+ val expected = breeze.linalg.cov(mat.toBreeze())
+ assert(closeToZero(abs(expected) - abs(result.toBreeze.asInstanceOf[BDM[Double]])))
+ }
+ }
+
+ test("covariance matrix is symmetric (SPARK-10875)") {
+ val rdd = RandomRDDs.normalVectorRDD(sc, 100, 10, 0, 0)
+ val matrix = new RowMatrix(rdd)
+ val cov = matrix.computeCovariance()
+ for (i <- 0 until cov.numRows; j <- 0 until i) {
+ assert(cov(i, j) === cov(j, i))
+ }
+ }
}
class RowMatrixClusterSuite extends SparkFunSuite with LocalClusterSparkContext {
--
GitLab