FastTree and LightGbm base class is specific to Ranking

The only training task that requires a RowGroupColumn is ranking. However, the base class of FastTree, FastForest and LightGbm requires it:

machinelearning/src/Microsoft.ML.Data/Training/TrainerEstimatorBase.cs

Line 176 in 70ef7ec

    
           public abstract class TrainerEstimatorBaseWithGroupId<TTransformer, TModel> : TrainerEstimatorBase<TTransformer, TModel>

The same is true for the options classes in FastTree, FastForest and LightGbm:

machinelearning/src/Microsoft.ML.Data/Training/TrainerInputBase.cs

Lines 98 to 111 in 70ef7ec

    
           /// <summary> 
        
           /// The base class for all trainer inputs that support a group column. 
        
           /// </summary> 
        
           [TlcModule.EntryPointKind(typeof(CommonInputs.ITrainerInputWithGroupId))] 
        
           public abstract class TrainerInputBaseWithGroupId : TrainerInputBaseWithWeight 
        
           { 
        
               private protected TrainerInputBaseWithGroupId() { } 
        
               /// <summary> 
        
               /// Column to use for example groupId. 
        
               /// </summary> 
        
               [Argument(ArgumentType.AtMostOnce, HelpText = "Column to use for example groupId", ShortName = "groupId", SortOrder = 5, Visibility = ArgumentAttribute.VisibilityType.EntryPointsOnly)] 
        
               public string RowGroupColumnName = null; 
        
           }

We should fix the class hierarchy so that the base classes become TrainerEstimatorBase instead of TrainerEstimatorBaseWithGroupId and TrainerInputBaseWithWeight instead of TrainerInputBaseWithGroupId. This is a breaking API change.

This related to #3365.

I am afraid that this will require a breaking API change at this point. This is something that we should consider fixing for 2.0. I am switching the label to keep track of these problems.

Triage: This looks like an old API-design debt. We currently do not plan to introduce breaking changes in this domain so I'll close this as not planned.

Navigation Menu

Repository navigation

FastTree and LightGbm base class is specific to Ranking #3556

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

	/// <summary>
	/// The base class for all trainer inputs that support a group column.
	/// </summary>
	[TlcModule.EntryPointKind(typeof(CommonInputs.ITrainerInputWithGroupId))]
	public abstract class TrainerInputBaseWithGroupId : TrainerInputBaseWithWeight
	{
	private protected TrainerInputBaseWithGroupId() { }

	/// <summary>
	/// Column to use for example groupId.
	/// </summary>
	[Argument(ArgumentType.AtMostOnce, HelpText = "Column to use for example groupId", ShortName = "groupId", SortOrder = 5, Visibility = ArgumentAttribute.VisibilityType.EntryPointsOnly)]
	public string RowGroupColumnName = null;
	}

FastTree and LightGbm base class is specific to Ranking #3556

Description

Activity

artidoro commented on May 29, 2019

matouskozak commented on Oct 23, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions