oneDAL Binary Classification LBFGS - Index was outside the bounds of the array.

[luisquintanilla]

System Information (please complete the following information):

• OS & Version: Windows 11
• ML.NET Version: ML.NET 3.0 prerelease
• .NET Version: .NET 6 & .NET 6

Describe the bug

Training a binary classification model using the Lbfgs trainer and setting the MLNET_BACKEND environment variable to ONEDAL produces the following error:

Unhandled exception. System.IndexOutOfRangeException: Index was outside the bounds of the array.
 at Microsoft.ML.Trainers.LbfgsTrainerBase`3.TrainCoreOneDal(IChannel ch, RoleMappedData data)
 at Microsoft.ML.Trainers.LbfgsTrainerBase`3.TrainModelCore(TrainContext context)
 at Microsoft.ML.Trainers.TrainerEstimatorBase`2.TrainTransformer(IDataView trainSet, IDataView validationSet, IPredictor initPredictor)
 at Microsoft.ML.Trainers.TrainerEstimatorBase`2.Fit(IDataView input)
at Microsoft.ML.Data.EstimatorChain`1.Fit(IDataView input)
at Program.<Main>$(String[] args) in C:\Dev\OneDalTest\OneDalTest\Program.cs:line 31

To Reproduce

1. Create a C# console application
2. Install the latest prerelease versions of Microsoft.ML and Microsoft.ML.OneDAL packages.
3. Paste the following code into the Program.cs file.

// Initialize MLContext
var ctx = new MLContext();

// Define data
var trainingData = new [] 
{
 new {Arch="ARM", Trainer="LightGBM", oneDALSupport=false},
 new {Arch="x86", Trainer="FastTree", oneDALSupport=true},
 new {Arch="x86", Trainer="LbfgsLogisticRegression", oneDALSupport=true},
 new {Arch="ARM", Trainer="FastTree", oneDALSupport=false}
};

// Load data into IDataView
var trainingDv = ctx.Data.LoadFromEnumerable(trainingData);

// Define data processing pipeline & trainer
var pipeline = 
 ctx.Transforms.Categorical.OneHotEncoding(new [] {
 new InputOutputColumnPair("ArchEncoded", "Arch"),
 new InputOutputColumnPair("TrainerEncoded", "Trainer")})
 .Append(ctx.Transforms.Concatenate("Features", "ArchEncoded", "TrainerEncoded"))
 .Append(ctx.BinaryClassification.Trainers.LbfgsLogisticRegression(labelColumnName:"oneDALSupport"));

// Train model
var model = pipeline.Fit(trainingDv);

4. Set the MLNET_BACKEND environment variable to ONEDAL
5. Run the application.

Expected behavior
The model trains successfully.

Additional context

The same code using the FastTree trainer trains the model successfully.

[michaelgsharp]

@rgesteve for visability.