Details
-
Bug
-
Status: Resolved
-
Major
-
Resolution: Cannot Reproduce
-
2.4.0
-
None
-
None
Description
RegressionMetrics fails in Spark 2.4 when running via Anaconda on a Windows machine. A java error comes back saying that "python worker failed to connect back". This makes all the evaluation metrics (https://spark.apache.org/docs/2.2.0/mllib-evaluation-metrics.html) unusable for scoring model performance.
Reverted to Spark 2.3 and did not have this issue, also tested 2.2 and did not have this issue. So it appears to be a bug specific to Spark 2.4. Likely also affects other evaluation metric types, e.g.BinaryClassificationMetrics.