[SPARK-31925] Summary.totalIterations greater than maxIters

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 2.4.6, 3.0.0
• Fix Version/s: 3.1.0
• Component/s: ML, PySpark
• Labels: release-notes
• Docs Text: Before Spark 3.1, if maxIter=n in LogisticRegression and LinearRegression, then the result summary's totalIterations returns n+1. It now correctly returns n, in 3.1. The objective history is unchanged.
• Assignee: Huaxin Gao
• Reporter: Ruifeng Zheng
• Created: 08/Jun/20 07:05
• Updated: 15/Jun/20 13:49
• Resolved: 15/Jun/20 13:49

Description

I am not sure whether it is a bug, but if we set maxIter=n in LiR/LiR/etc, the model.summary.totalIterations will return n+1 if the training procedure does not drop out.

 

friendly ping huaxingao

Issue Links

• links to: [Github] Pull Request #28786 (huaxingao)