[SPARK-15169] Consider improving HasSolver to allow generalization

• Type: Improvement
• Status: Resolved
• Priority: Trivial
• Resolution: Incomplete
• Component/s: ML
• Labels: bulk-closed
• Assignee: Unassigned
• Reporter: Holden Karau
• Created: 06/May/16 00:29
• Updated: 21/May/19 04:37
• Resolved: 21/May/19 04:37

Description

The current HasSolver shared param has a fixed default value of "auto" and no validation. Some algorithms (see `MultilayerPerceptronClassifier`) have different default values or validators. This results in either a mostly duplicated param (as in `MultilayerPerceptronClassifier`) or incorrect scaladoc (as in `GeneralizedLinearRegression`).