Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-26387

Parallelism seems to cause difference in CrossValidation model metrics

    XMLWordPrintableJSON

    Details

    • Type: Bug
    • Status: Resolved
    • Priority: Major
    • Resolution: Not A Problem
    • Affects Version/s: 2.3.1, 2.3.2
    • Fix Version/s: None
    • Component/s: ML, MLlib
    • Labels:
      None

      Description

      I can only reproduce this issue when running Spark on different Amazon EMR versions, but it seems that between Spark 2.3.1 and 2.3.2 (corresponding to EMR versions 5.17/5.18) the presence of the parallelism parameter was causing AUC metric to increase. Literally, I run the same exact code with and without parallelism and the AUC of my models (logistic regression) are changing significantly. I can't find a previous bug report relating to this, so I'm posting this as new.

        Attachments

          Activity

            People

            • Assignee:
              Unassigned
              Reporter:
              zamir.evan@gmail.com Evan Zamir
            • Votes:
              0 Vote for this issue
              Watchers:
              2 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved: