Schema mismatch for label column 'Label': expected Single, got Vector<Single> (Parameter 'labelCol')

[ooples]

I'm trying out this nuget package and experimenting with the different normalization options for regression and prediction trainers. My code is working perfectly when I use Single values (float) but I saw that there were bunch of normalization options that only seemed to work using a vector of Single values so I get the error when I change all of my float values to a float array. Am I just missing something obvious?

FYI I'm using ML.NET 3.0.0-preview.23266.6 for this example

var trainingCount = 50;
var modelInputList = new List<ModelInput>();
var estCount = valuesList.Any() ? valuesList.First().ValueList.Count : 0;

for (int j = 0; j < estCount; j++)
{
    var modelInput = new ModelInput();

    var actual = j < estCount - 1 ? actualList[j] : 0;
    modelInput.Actual = new float[] { Convert.ToSingle(actual) };

    for (int k = 0; k < valueCount; k++)
    {
        var rvItem = Convert.ToSingle(valuesList[k].ValueList[j]);

        switch (k)
        {
            case 0:
                modelInput.Input1 = new float[] { rvItem };
                break;
            case 1:
                modelInput.Input2 = new float[] { rvItem };
                break;
            case 2:
                modelInput.Input3 = new float[] { rvItem };
                break;
            default:
                break;
        }
    }

    modelInputList.Add(modelInput);
}

var firstHalf = mlContext.Data.LoadFromEnumerable(modelInputList.Take(trainingCount));
var secondHalf = mlContext.Data.LoadFromEnumerable(modelInputList.Skip(trainingCount));
var dataProcessPipeline = mlContext.Transforms
                    .CopyColumns("Label", nameof(ModelInput.Actual))
                    .Append(mlContext.Transforms.NormalizeBinning(outputColumnName: nameof(ModelInput.Input1)))
                    .Append(mlContext.Transforms.NormalizeBinning(outputColumnName: nameof(ModelInput.Input2)))
                    .Append(mlContext.Transforms.NormalizeBinning(outputColumnName: nameof(ModelInput.Input3)))
                    .Append(mlContext.Transforms.Concatenate("Features", nameof(ModelInput.Input1),
                        nameof(ModelInput.Input2), nameof(ModelInput.Input3)));
var trainer = mlContext.Regression.Trainers.OnlineGradientDescent();
var trainingPipeline = dataProcessPipeline.Append(trainer);
var trainedModel = trainingPipeline.Fit(firstHalf); // getting the exception here
var trainingData = trainedModel.Transform(firstHalf);
var predictions = trainedModel.Transform(secondHalf);

public class ModelInput
{
    [LoadColumn(0)]
    [VectorType(1)]
    public float[] Actual { get; set; }

    [LoadColumn(1)]
    [VectorType(1)]
    public float[] Input1 { get; set; }

    [LoadColumn(2)]
    [VectorType(1)]
    public float[] Input2 { get; set; }

    [LoadColumn(3)]
    [VectorType(1)]
    public float[] Input3 { get; set; }
}

[rosebyte]

Regression trainers require the label column to be a scalar Single. Keep Actual as float; only the feature columns should be vectors where needed. The exception correctly reports the incompatible schema, so this is a usage issue rather than a defect.