[SPARK-2478] Add Python APIs for decision tree

• Type: New Feature
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Fix Version/s: 1.1.0
• Component/s: MLlib, PySpark
• Target Version/s: 1.1.0
• Assignee: Joseph K. Bradley
• Reporter: Xiangrui Meng
• Created: 14/Jul/14 21:22
• Updated: 02/Aug/14 20:08
• Resolved: 02/Aug/14 20:08

Description

In v1.0, we only support decision tree in Scala/Java. It would be nice to add Python support. It may require some refactoring of the current decision tree API to make it easier to construct a decision tree algorithm in Python.

1. Simplify decision tree constructors such that only simple types are used.
a. Hide the implementation of Impurity from users.
b. Replace enums by strings.
2. Make separate public decision tree classes for regression & classification (with shared internals). Eliminate algo parameter.
3. Implement wrappers in Python for DecisionTree.
4. Implement wrappers in Python for DecisionTreeModel.

Issue Links

• is related to: SPARK-2197 Spark invoke DecisionTree by Java
• links to: [Github] Pull Request #1727 (jkbradley)