[SPARK-28440][MLLIB][TEST] Use TestingUtils to compare floating point…

… values ## What changes were proposed in this pull request? Use `org.apache.spark.mllib.util.TestingUtils` object across `MLLIB` component to compare floating point values in tests. ## How was this patch tested? `build/mvn test` - existing tests against updated code. Closes apache#25191 from eugen-prokhorenko/mllib-testingutils-double-comparison. Authored-by: Ievgen Prokhorenko <[email protected]> Signed-off-by: Dongjoon Hyun <[email protected]>
koeninger · Jul 19, 2019 · 52ddf03 · 52ddf03
1 parent 127bc89
commit 52ddf03
Show file tree

Hide file tree

Showing 9 changed files with 52 additions and 46 deletions.
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/evaluation/MultilabelMetricsSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/evaluation/MultilabelMetricsSuite.scala
@@ -19,6 +19,7 @@ package org.apache.spark.mllib.evaluation
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.TestingUtils._
 import org.apache.spark.rdd.RDD
 
 class MultilabelMetricsSuite extends SparkFunSuite with MLlibTestSparkContext {
@@ -79,24 +80,24 @@ class MultilabelMetricsSuite extends SparkFunSuite with MLlibTestSparkContext {
     val hammingLoss = (1.0 / (7 * 3)) * (2 + 2 + 1 + 0 + 0 + 1 + 1)
     val strictAccuracy = 2.0 / 7
     val accuracy = 1.0 / 7 * (1.0 / 3 + 1.0 /3 + 0 + 1.0 / 1 + 2.0 / 2 + 2.0 / 3 + 1.0 / 2)
-    assert(math.abs(metrics.precision(0.0) - precision0) < delta)
-    assert(math.abs(metrics.precision(1.0) - precision1) < delta)
-    assert(math.abs(metrics.precision(2.0) - precision2) < delta)
-    assert(math.abs(metrics.recall(0.0) - recall0) < delta)
-    assert(math.abs(metrics.recall(1.0) - recall1) < delta)
-    assert(math.abs(metrics.recall(2.0) - recall2) < delta)
-    assert(math.abs(metrics.f1Measure(0.0) - f1measure0) < delta)
-    assert(math.abs(metrics.f1Measure(1.0) - f1measure1) < delta)
-    assert(math.abs(metrics.f1Measure(2.0) - f1measure2) < delta)
-    assert(math.abs(metrics.microPrecision - microPrecisionClass) < delta)
-    assert(math.abs(metrics.microRecall - microRecallClass) < delta)
-    assert(math.abs(metrics.microF1Measure - microF1MeasureClass) < delta)
-    assert(math.abs(metrics.precision - macroPrecisionDoc) < delta)
-    assert(math.abs(metrics.recall - macroRecallDoc) < delta)
-    assert(math.abs(metrics.f1Measure - macroF1MeasureDoc) < delta)
-    assert(math.abs(metrics.hammingLoss - hammingLoss) < delta)
-    assert(math.abs(metrics.subsetAccuracy - strictAccuracy) < delta)
-    assert(math.abs(metrics.accuracy - accuracy) < delta)
+    assert(metrics.precision(0.0) ~== precision0 absTol delta)
+    assert(metrics.precision(1.0) ~== precision1 absTol delta)
+    assert(metrics.precision(2.0) ~== precision2 absTol delta)
+    assert(metrics.recall(0.0) ~== recall0 absTol delta)
+    assert(metrics.recall(1.0) ~== recall1 absTol delta)
+    assert(metrics.recall(2.0) ~== recall2 absTol delta)
+    assert(metrics.f1Measure(0.0) ~== f1measure0 absTol delta)
+    assert(metrics.f1Measure(1.0) ~== f1measure1 absTol delta)
+    assert(metrics.f1Measure(2.0) ~== f1measure2 absTol delta)
+    assert(metrics.microPrecision ~== microPrecisionClass absTol delta)
+    assert(metrics.microRecall ~== microRecallClass absTol delta)
+    assert(metrics.microF1Measure ~== microF1MeasureClass absTol delta)
+    assert(metrics.precision ~== macroPrecisionDoc absTol delta)
+    assert(metrics.recall ~== macroRecallDoc absTol delta)
+    assert(metrics.f1Measure ~== macroF1MeasureDoc absTol delta)
+    assert(metrics.hammingLoss ~== hammingLoss absTol delta)
+    assert(metrics.subsetAccuracy ~== strictAccuracy absTol delta)
+    assert(metrics.accuracy ~== accuracy absTol delta)
     assert(metrics.labels.sameElements(Array(0.0, 1.0, 2.0)))
   }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/fpm/AssociationRulesSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/fpm/AssociationRulesSuite.scala
@@ -18,6 +18,7 @@ package org.apache.spark.mllib.fpm
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.TestingUtils._
 
 class AssociationRulesSuite extends SparkFunSuite with MLlibTestSparkContext {
 
@@ -63,7 +64,7 @@ class AssociationRulesSuite extends SparkFunSuite with MLlibTestSparkContext {
        [1] 23
      */
     assert(results1.size === 23)
-    assert(results1.count(rule => math.abs(rule.confidence - 1.0D) < 1e-6) == 23)
+    assert(results1.count(rule => rule.confidence ~= 1.0D absTol 1e-6) == 23)
 
     val results2 = ar
       .setMinConfidence(0)
@@ -84,7 +85,7 @@ class AssociationRulesSuite extends SparkFunSuite with MLlibTestSparkContext {
        [1] 23
      */
     assert(results2.size === 30)
-    assert(results2.count(rule => math.abs(rule.confidence - 1.0D) < 1e-6) == 23)
+    assert(results2.count(rule => rule.confidence ~= 1.0D absTol 1e-6) == 23)
   }
 }
 
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/fpm/FPGrowthSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/fpm/FPGrowthSuite.scala
@@ -18,6 +18,7 @@ package org.apache.spark.mllib.fpm
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.TestingUtils._
 import org.apache.spark.util.Utils
 
 class FPGrowthSuite extends SparkFunSuite with MLlibTestSparkContext {
@@ -172,7 +173,7 @@ class FPGrowthSuite extends SparkFunSuite with MLlibTestSparkContext {
       .collect()
 
     assert(rules.size === 23)
-    assert(rules.count(rule => math.abs(rule.confidence - 1.0D) < 1e-6) == 23)
+    assert(rules.count(rule => rule.confidence ~= 1.0D absTol 1e-6) == 23)
   }
 
   test("FP-Growth using Int type") {

diff --git a/mllib/src/test/scala/org/apache/spark/mllib/linalg/distributed/IndexedRowMatrixSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/linalg/distributed/IndexedRowMatrixSuite.scala
@@ -22,6 +22,7 @@ import breeze.linalg.{diag => brzDiag, DenseMatrix => BDM, DenseVector => BDV}
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.linalg._
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.TestingUtils._
 import org.apache.spark.rdd.RDD
 
 class IndexedRowMatrixSuite extends SparkFunSuite with MLlibTestSparkContext {
@@ -238,7 +239,7 @@ class IndexedRowMatrixSuite extends SparkFunSuite with MLlibTestSparkContext {
 
     for (i <- 0 until n; j <- i + 1 until n) {
       val trueResult = gram(i, j) / scala.math.sqrt(gram(i, i) * gram(j, j))
-      assert(math.abs(G(i, j) - trueResult) < 1e-6)
+      assert(G(i, j) ~== trueResult absTol 1e-6)
     }
   }
 

diff --git a/mllib/src/test/scala/org/apache/spark/mllib/random/RandomDataGeneratorSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/random/RandomDataGeneratorSuite.scala
@@ -20,9 +20,9 @@ package org.apache.spark.mllib.random
 import org.apache.commons.math3.special.Gamma
 
 import org.apache.spark.SparkFunSuite
+import org.apache.spark.mllib.util.TestingUtils._
 import org.apache.spark.util.StatCounter
 
-// TODO update tests to use TestingUtils for floating point comparison after PR 1367 is merged
 class RandomDataGeneratorSuite extends SparkFunSuite {
 
   def apiChecks(gen: RandomDataGenerator[Double]) {
@@ -61,8 +61,8 @@ class RandomDataGeneratorSuite extends SparkFunSuite {
       gen.setSeed(seed.toLong)
       val sample = (0 until 100000).map { _ => gen.nextValue()}
       val stats = new StatCounter(sample)
-      assert(math.abs(stats.mean - mean) < epsilon)
-      assert(math.abs(stats.stdev - stddev) < epsilon)
+      assert(stats.mean ~== mean absTol epsilon)
+      assert(stats.stdev ~== stddev absTol epsilon)
     }
   }
 

diff --git a/mllib/src/test/scala/org/apache/spark/mllib/random/RandomRDDsSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/random/RandomRDDsSuite.scala
@@ -23,14 +23,13 @@ import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.linalg.Vector
 import org.apache.spark.mllib.rdd.{RandomRDD, RandomRDDPartition}
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.TestingUtils._
 import org.apache.spark.rdd.RDD
 import org.apache.spark.util.StatCounter
 
 /*
  * Note: avoid including APIs that do not set the seed for the RNG in unit tests
  * in order to guarantee deterministic behavior.
- *
- * TODO update tests to use TestingUtils for floating point comparison after PR 1367 is merged
  */
 class RandomRDDsSuite extends SparkFunSuite with MLlibTestSparkContext with Serializable {
 
@@ -43,8 +42,8 @@ class RandomRDDsSuite extends SparkFunSuite with MLlibTestSparkContext with Seri
     val stats = rdd.stats()
     assert(expectedSize === stats.count)
     assert(expectedNumPartitions === rdd.partitions.size)
-    assert(math.abs(stats.mean - expectedMean) < epsilon)
-    assert(math.abs(stats.stdev - expectedStddev) < epsilon)
+    assert(stats.mean ~== expectedMean absTol epsilon)
+    assert(stats.stdev ~== expectedStddev absTol epsilon)
   }
 
   // assume test RDDs are small
@@ -63,8 +62,8 @@ class RandomRDDsSuite extends SparkFunSuite with MLlibTestSparkContext with Seri
     }}
     assert(expectedRows === values.size / expectedColumns)
     val stats = new StatCounter(values)
-    assert(math.abs(stats.mean - expectedMean) < epsilon)
-    assert(math.abs(stats.stdev - expectedStddev) < epsilon)
+    assert(stats.mean ~== expectedMean absTol epsilon)
+    assert(stats.stdev ~== expectedStddev absTol epsilon)
   }
 
   test("RandomRDD sizes") {

diff --git a/mllib/src/test/scala/org/apache/spark/mllib/stat/CorrelationSuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/stat/CorrelationSuite.scala
@@ -26,6 +26,7 @@ import org.apache.spark.mllib.random.RandomRDDs
 import org.apache.spark.mllib.stat.correlation.{Correlations, PearsonCorrelation,
   SpearmanCorrelation}
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.TestingUtils._
 
 class CorrelationSuite extends SparkFunSuite with MLlibTestSparkContext with Logging {
 
@@ -57,15 +58,15 @@ class CorrelationSuite extends SparkFunSuite with MLlibTestSparkContext with Log
     val expected = 0.6546537
     val default = Statistics.corr(x, y)
     val p1 = Statistics.corr(x, y, "pearson")
-    assert(approxEqual(expected, default))
-    assert(approxEqual(expected, p1))
+    assert(expected ~== default absTol 1e-6)
+    assert(expected ~== p1 absTol 1e-6)
 
     // numPartitions >= size for input RDDs
     for (numParts <- List(xData.size, xData.size * 2)) {
       val x1 = sc.parallelize(xData, numParts)
       val y1 = sc.parallelize(yData, numParts)
       val p2 = Statistics.corr(x1, y1)
-      assert(approxEqual(expected, p2))
+      assert(expected ~== p2 absTol 1e-6)
     }
 
     // RDD of zero variance
@@ -78,14 +79,14 @@ class CorrelationSuite extends SparkFunSuite with MLlibTestSparkContext with Log
     val y = sc.parallelize(yData)
     val expected = 0.5
     val s1 = Statistics.corr(x, y, "spearman")
-    assert(approxEqual(expected, s1))
+    assert(expected ~== s1 absTol 1e-6)
 
     // numPartitions >= size for input RDDs
     for (numParts <- List(xData.size, xData.size * 2)) {
       val x1 = sc.parallelize(xData, numParts)
       val y1 = sc.parallelize(yData, numParts)
       val s2 = Statistics.corr(x1, y1, "spearman")
-      assert(approxEqual(expected, s2))
+      assert(expected ~== s2 absTol 1e-6)
     }
 
     // RDD of zero variance => zero variance in ranks
@@ -141,14 +142,14 @@ class CorrelationSuite extends SparkFunSuite with MLlibTestSparkContext with Log
     val a = RandomRDDs.normalRDD(sc, 100000, 10).map(_ + 1000000000.0)
     val b = RandomRDDs.normalRDD(sc, 100000, 10).map(_ + 1000000000.0)
     val p = Statistics.corr(a, b, method = "pearson")
-    assert(approxEqual(p, 0.0, 0.01))
+    assert(p ~== 0.0 absTol 0.01)
   }
 
   def approxEqual(v1: Double, v2: Double, threshold: Double = 1e-6): Boolean = {
     if (v1.isNaN) {
       v2.isNaN
     } else {
-      math.abs(v1 - v2) <= threshold
+      v1 ~== v2 absTol threshold
     }
   }
 

diff --git a/mllib/src/test/scala/org/apache/spark/mllib/stat/KernelDensitySuite.scala b/mllib/src/test/scala/org/apache/spark/mllib/stat/KernelDensitySuite.scala
@@ -21,6 +21,7 @@ import org.apache.commons.math3.distribution.NormalDistribution
 
 import org.apache.spark.SparkFunSuite
 import org.apache.spark.mllib.util.MLlibTestSparkContext
+import org.apache.spark.mllib.util.TestingUtils._
 
 class KernelDensitySuite extends SparkFunSuite with MLlibTestSparkContext {
   test("kernel density single sample") {
@@ -29,8 +30,8 @@ class KernelDensitySuite extends SparkFunSuite with MLlibTestSparkContext {
     val densities = new KernelDensity().setSample(rdd).setBandwidth(3.0).estimate(evaluationPoints)
     val normal = new NormalDistribution(5.0, 3.0)
     val acceptableErr = 1e-6
-    assert(math.abs(densities(0) - normal.density(5.0)) < acceptableErr)
-    assert(math.abs(densities(1) - normal.density(6.0)) < acceptableErr)
+    assert(densities(0) ~== normal.density(5.0) absTol acceptableErr)
+    assert(densities(1) ~== normal.density(6.0) absTol acceptableErr)
   }
 
   test("kernel density multiple samples") {
@@ -40,9 +41,9 @@ class KernelDensitySuite extends SparkFunSuite with MLlibTestSparkContext {
     val normal1 = new NormalDistribution(5.0, 3.0)
     val normal2 = new NormalDistribution(10.0, 3.0)
     val acceptableErr = 1e-6
-    assert(math.abs(
-      densities(0) - (normal1.density(5.0) + normal2.density(5.0)) / 2) < acceptableErr)
-    assert(math.abs(
-      densities(1) - (normal1.density(6.0) + normal2.density(6.0)) / 2) < acceptableErr)
+    assert(
+      densities(0) ~== ((normal1.density(5.0) + normal2.density(5.0)) / 2) absTol acceptableErr)
+    assert(
+      densities(1) ~== ((normal1.density(6.0) + normal2.density(6.0)) / 2) absTol acceptableErr)
   }
 }
diff --git a/mllib/src/test/scala/org/apache/spark/mllib/tree/EnsembleTestHelper.scala b/mllib/src/test/scala/org/apache/spark/mllib/tree/EnsembleTestHelper.scala
@@ -22,6 +22,7 @@ import scala.collection.mutable
 import org.apache.spark.mllib.linalg.Vectors
 import org.apache.spark.mllib.regression.LabeledPoint
 import org.apache.spark.mllib.tree.model.TreeEnsembleModel
+import org.apache.spark.mllib.util.TestingUtils._
 import org.apache.spark.util.StatCounter
 
 object EnsembleTestHelper {
@@ -43,8 +44,8 @@ object EnsembleTestHelper {
       values ++= row
     }
     val stats = new StatCounter(values)
-    assert(math.abs(stats.mean - expectedMean) < epsilon)
-    assert(math.abs(stats.stdev - expectedStddev) < epsilon)
+    assert(stats.mean ~== expectedMean absTol epsilon)
+    assert(stats.stdev ~== expectedStddev absTol epsilon)
   }
 
   def validateClassifier(