Github user srowen commented on a diff in the pull request: https://github.com/apache/spark/pull/1425#discussion_r14988046 --- Diff: mllib/src/test/scala/org/apache/spark/mllib/evaluation/BinaryClassificationMetricsSuite.scala --- @@ -20,8 +20,20 @@ package org.apache.spark.mllib.evaluation import org.scalatest.FunSuite import org.apache.spark.mllib.util.LocalSparkContext +import org.apache.spark.mllib.util.TestingUtils._ class BinaryClassificationMetricsSuite extends FunSuite with LocalSparkContext { + + implicit class SeqDoubleWithAlmostEquals(val x: Seq[Double]) { + def almostEquals(y: Seq[Double], eps: Double = 1E-6): Boolean = --- End diff -- 1.0e-6 is way bigger than an ulp for a double; 1.0e-12 is more like it. I understand a complex calculation might legitimately vary by significantly more than an ulp depending on the implementation. As @mengxr says where you mean to allow significantly more than machine precision worth of noise, that's probably good to do with an explicitly larger epsilon. But this is certainly a good step forward already.
--- 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 infrastruct...@apache.org or file a JIRA ticket with INFRA. ---