[SPARK-31169] Random Forest in SparkML 2.3.3 vs 2.4.x

• Type: Question
• Status: Resolved
• Priority: Major
• Resolution: Invalid
• Affects Version/s: 2.3.3, 2.4.0, 2.4.3
• Component/s: ML
• Labels: MLLib, RandomForest SparkML
• Assignee: Unassigned
• Reporter: Nguyen Nhanduc
• Created: 17/Mar/20 01:24
• Updated: 12/Dec/22 18:10
• Resolved: 23/Mar/20 05:08

Description

Hi all,

When I trained the model with the Random Forest algorithm, I got different results in different versions of spark, the same input, label ratio, hyperparameter for all training. Detailed training results in the attached file. Model training results with spark 2.3.3 are much better, so I want to ask if there have been any changes to the random forest (or other algorithms) in mllib?

Many thanks.