[SPARK-17133] Improvements to linear methods in Spark

• Type: Umbrella
• Status: Resolved
• Priority: Major
• Resolution: Incomplete
• Component/s: ML, MLlib
• Labels: bulk-closed
• Assignee: Unassigned
• Reporter: Seth Hendrickson
• Created: 18/Aug/16 17:50
• Updated: 21/May/19 04:10
• Resolved: 21/May/19 04:10

Description

This JIRA is for tracking several improvements that we should make to Linear/Logistic regression in Spark.

Issue Links

• is related to: SPARK-7159 Support multiclass logistic regression in spark.ml
• is related to: SPARK-16592 Improving ml.Logistic Regression on speed and scalability
• links to: [Github] Pull Request #19072 (WeichenXu123)

Sub-Tasks

• Use level 2 BLAS operations in LogisticAggregator: Resolved
• Make tree aggregation level in linear/logistic regression configurable: Resolved
• Consolidate code in linear/logistic regression where possible: Resolved
• Design optimizer interface for ML algorithms: Resolved
• Add compressed support for multinomial logistic regression coefficients: Resolved
• Python API for multinomial logistic regression: Resolved
• Add model summary for MultinomialLogisticRegression: Resolved
• Add initial model to MultinomialLogisticRegression: Resolved
• Decide how to handle inferring number of classes in Multinomial logistic regression: Resolved
• Merge MLOR into a single LOR interface: Resolved
• Investigate numerical instability for MLOR without regularization: Resolved
• Proper handling for unseen labels in logistic regression training.: Resolved
• Avoid unnecessary standardization in multinomial logistic regression training: Resolved
• Use level 3 BLAS operations in LogisticAggregator: Resolved
• Resolve code duplication for classification/regression summarizers: Resolved