Github user imatiach-msft commented on a diff in the pull request:
https://github.com/apache/spark/pull/16722#discussion_r99279066
--- Diff:
mllib/src/test/scala/org/apache/spark/mllib/tree/ImpuritySuite.scala ---
@@ -18,23 +18,62 @@
package org.apache.spark.mllib.tree
import org.apache.spark.SparkFunSuite
-import org.apache.spark.mllib.tree.impurity.{EntropyAggregator,
GiniAggregator}
+import org.apache.spark.ml.util.TestingUtils._
+import org.apache.spark.mllib.tree.impurity._
/**
* Test suites for [[GiniAggregator]] and [[EntropyAggregator]].
*/
class ImpuritySuite extends SparkFunSuite {
+
+ private val seed = 42
+
test("Gini impurity does not support negative labels") {
val gini = new GiniAggregator(2)
intercept[IllegalArgumentException] {
- gini.update(Array(0.0, 1.0, 2.0), 0, -1, 0.0)
+ gini.update(Array(0.0, 1.0, 2.0), 0, -1, 3, 0.0)
}
}
test("Entropy does not support negative labels") {
val entropy = new EntropyAggregator(2)
intercept[IllegalArgumentException] {
- entropy.update(Array(0.0, 1.0, 2.0), 0, -1, 0.0)
+ entropy.update(Array(0.0, 1.0, 2.0), 0, -1, 3, 0.0)
+ }
+ }
+
+ test("Classification impurities are insensitive to scaling") {
+ val rng = new scala.util.Random(seed)
+ val weightedCounts = Array.fill(5)(rng.nextDouble())
+ val smallWeightedCounts = weightedCounts.map(_ * 0.0001)
+ val largeWeightedCounts = weightedCounts.map(_ * 10000)
+ Seq(Gini, Entropy).foreach { impurity =>
+ val impurity1 = impurity.calculate(weightedCounts,
weightedCounts.sum)
+ assert(impurity.calculate(smallWeightedCounts,
smallWeightedCounts.sum)
+ ~== impurity1 relTol 0.005)
+ assert(impurity.calculate(largeWeightedCounts,
largeWeightedCounts.sum)
+ ~== impurity1 relTol 0.005)
}
}
+ test("Regression impurities are insensitive to scaling") {
+ def computeStats(samples: Seq[Double], weights: Seq[Double]): (Double,
Double, Double) = {
+ samples.zip(weights).foldLeft((0.0, 0.0, 0.0)) { case ((wn, wy,
wyy), (y, w)) =>
+ (wn + w, wy + w * y, wyy + w * y * y)
+ }
+ }
+ val rng = new scala.util.Random(seed)
+ val samples = Array.fill(10)(rng.nextDouble())
+ val _weights = Array.fill(10)(rng.nextDouble())
+ val smallWeights = _weights.map(_ * 0.0001)
+ val largeWeights = _weights.map(_ * 10000)
+ val (count, sum, sumSquared) = computeStats(samples, _weights)
+ Seq(Variance).foreach { impurity =>
+ val impurity1 = impurity.calculate(count, sum, sumSquared)
+ val (smallCount, smallSum, smallSumSquared) = computeStats(samples,
smallWeights)
+ val (largeCount, largeSum, largeSumSquared) = computeStats(samples,
largeWeights)
+ assert(impurity.calculate(smallCount, smallSum, smallSumSquared) ~==
impurity1 relTol 0.005)
+ assert(impurity.calculate(largeCount, largeSum, largeSumSquared) ~==
impurity1 relTol 0.005)
--- End diff --
these are really nice tests
---
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 [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
