[SPARK-4875] Separate Transformer, Estimator params

• Type: Improvement
• Status: Closed
• Priority: Minor
• Resolution: Duplicate
• Affects Version/s: 1.2.0
• Component/s: ML
• Target Version/s: 1.3.0
• Assignee: Unassigned
• Reporter: Joseph K. Bradley
• Created: 17/Dec/14 19:33
• Updated: 17/Dec/14 22:21
• Resolved: 17/Dec/14 22:21

Description

Currently, in spark.ml, an Estimator and its Model (Transformer) both use the same set of parameters. (E.g., LogisticRegression) The problem is that some Estimator parameters are not relevant for Transformers, so it is non-sensical to include them in the Transformers.

Proposal: When a parameter is only relevant to the Estimator, only include it in the Estimator. In these cases, if there is a separate parameter trait (e.g., LogisticRegressionParams), that trait should be separated into a model params trait and an estimator params trait which inherits from the model params:

trait LogisticRegressionModelParams
trait LogisticRegressionParams extends LogisticRegressionModelParams

In cases where the 2 classes use the same set of parameters, then they may share the same params trait.

Issue Links

• duplicates: SPARK-4766 ML Estimator Params should be distinct from Transformer Params