Can't get KMeans clustering to work in Microsoft.ml (C#)

Question

Below is a simple demonstration of my issue. When I execute "Evaluate" I get an exception stating Schema mismatch for score column 'Score': expected known-size vector of Single, got Vector<Single> The tutorial use a unknown-size vector so it should work? https://learn.microsoft.com/en-us/dotnet/machine-learning/tutorials/iris-clustering

If I try with a fixed size score (decorating with a [VectorType(3)] attribute) I instead get an Operation is not valid due to the current state of the object exception (I assume it should be 3 since I ask for three clusters and the score should be three distances)

What am I doing wrong?

public class SampleInputModel
{
    [VectorType(3)]
    public float SimpleVector { get; set; }

    [ColumnName("PredictedLabel")]
    public uint PredictedClusterId;

    [ColumnName("Score")]
    public float[] Distances;
}

   [Fact]
    public void Sample()
    {
        var mlContext = new MLContext(seed: 1);

        var random = new Random().NextDoubleSequence();
        var dummyData = new SampleInputModel[100];
        for (int i = 0; i < 100; i++)
        {
            dummyData[i] = new SampleInputModel { SimpleVector = new float[] { (float)random.ElementAt(i), (float)random.ElementAt(i), (float)random.ElementAt(i) } };
        }
        var trainingData = mlContext.Data.LoadFromEnumerable<SampleInputModel>(dummyData.Take(50));
        var pipeline = mlContext.Transforms.Concatenate("Features", "SimpleVector");

        pipeline.Append(mlContext.Clustering.Trainers.KMeans("Features", numberOfClusters: 3));
        var trainedModel = pipeline.Fit(trainingData);

        var testData = mlContext.Data.LoadFromEnumerable<SampleInputModel>(dummyData.TakeLast(50));
        var predictions = trainedModel.Transform(testData);
        var metrics = mlContext.Clustering.Evaluate(predictions, scoreColumnName: "Score", featureColumnName: "Features");
    }

Answer 1

I solved it!

Seems like KMeans models doesn't like concatination of features (like regression/classification models do).

Suppying the feature vector explicitly works fine:

    [Fact]
    public void Sample()
    {
        var mlContext = new MLContext(seed: 1);

        var random = new Random().NextDoubleSequence();
        var dummyData = new SampleInputModel[100];
        for (int i = 0; i < 100; i++)
        {
            dummyData[i] = new SampleInputModel { SimpleVector = new float[] { (float)random.ElementAt(i), (float)random.ElementAt(i), (float)random.ElementAt(i) } };
        }
        var trainingData = mlContext.Data.LoadFromEnumerable<SampleInputModel>(dummyData.Take(50));
        var pipeline = mlContext.Clustering.Trainers.KMeans("SimpleVector", numberOfClusters: 3);
        var trainedModel = pipeline.Fit(trainingData);

        var testData = mlContext.Data.LoadFromEnumerable<SampleInputModel>(dummyData.TakeLast(50));
        var predictions = trainedModel.Transform(testData);

        var metrics = mlContext.Clustering.Evaluate(predictions, null, "Score", "SimpleVector");
    }