[SPARK-18481] ML 2.1 QA: Remove deprecated methods for ML

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 2.1.0
• Component/s: ML, PySpark
• Assignee: Yanbo Liang
• Reporter: Yanbo Liang
• Created: 17/Nov/16 04:05
• Updated: 30/Nov/16 22:50
• Resolved: 26/Nov/16 13:30

Description

Remove deprecated methods for ML.

This task removed the following (deprecated) public APIs in org.apache.spark.ml:

• classification.RandomForestClassificationModel.numTrees (This now refers to the Param called "numTrees")
• feature.ChiSqSelectorModel.setLabelCol
• regression.LinearRegressionSummary.model
• regression.RandomForestRegressionModel.numTrees (This now refers to the Param called "numTrees")
• PipelineStage.validateParams
• Evaluator.validateParams

This task made the following changes to match existing patterns for Params:

• These methods were made final:
  • classification.RandomForestClassificationModel.getNumTrees
  • regression.RandomForestRegressionModel.getNumTrees
• These methods return the concrete class type, rather than an arbitrary trait. This only affected Java compatibility, not Scala.
  • classification.RandomForestClassificationModel.setFeatureSubsetStrategy
  • regression.RandomForestRegressionModel.setFeatureSubsetStrategy

Issue Links

• Is contained by: SPARK-18318 ML, Graph 2.1 QA: API: New Scala APIs, docs
• is related to: SPARK-18592 Move DT/RF/GBT Param setter methods to subclasses
• links to: [Github] Pull Request #15913 (yanboliang)