[SPARK-10578] pyspark.ml.classification.RandomForestClassifer does not return `rawPrediction` column

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.4.0, 1.4.1
• Fix Version/s: 1.5.0
• Component/s: ML

Environment

CentOS, PySpark 1.4.1, Scala 2.10

• Flags: Important
• Assignee: Joseph K. Bradley
• Reporter: Karen Y. Ng
• Created: 12/Sep/15 23:10
• Updated: 14/Sep/15 17:06
• Resolved: 14/Sep/15 17:02

Description

To use `pyspark.ml.classification.RandomForestClassifer` with `BinaryClassificationEvaluator`, a column called `rawPrediction` needs to be returned by the `RandomForestClassifer`.
The PySpark documentation example of `logisticsRegression`outputs the `rawPrediction` column but not `RandomForestClassifier`.

Therefore, one is unable to use `RandomForestClassifier` with the evaluator nor put it in a pipeline with cross validation.

A relevant piece of code showing how to reproduce the bug can be found at:
https://gist.github.com/karenyyng/cf61ae655b032f754bfb

A relevant post due to this possible bug can also be found at:
http://apache-spark-user-list.1001560.n3.nabble.com/Issue-with-running-CrossValidator-with-RandomForestClassifier-on-dataset-td23791.html

Issue Links

• duplicates: SPARK-9447 Python RandomForestClassifier probabilityCol, rawPredictionCol