Onnx with single float output throws exception during MLContext predict

[xqiu]

System Information (please complete the following information):

• OS & Version: [Windows 11]
• ML.NET Version: [ML.NET v3.0.1]
• .NET Version: [.NET 8.0]
• Microsoft.ML.OnnxRuntime Version: [1.1.9.1]
• Microsoft.ML.OnnxTransformer Version: [1.1.9.1]

Describe the bug
I have an onnx file with only a single float32 output. By using the following class to map the output with mlContext.Transforms.ApplyOnnxModel, and CreatePredictionEngine, predict function will report System.ArgumentNullException: Value cannot be null. (Parameter 'source') at Microsoft.ML.Transforms.Onnx.OnnxTransformer.Mapper.<>c__DisplayClass16_0`1.

I can use InferenceSession to run it without problem.

To Reproduce
Steps to reproduce the behavior:

1. Clone https://github.com/xqiu/ml_onnx_output_error
2. Run the program to see the bug appearance

Code is:

    public class ModelInput
    {
        [ColumnName("input")]
        [VectorType(1, 1, 40, 40)]  // Adjust dimensions to match your model input
        public float[] Features { get; set; } = new float[1 * 1 * 40 * 40];  // Flattened 4D array for ONNX input
    }

    public class ModelOutput
    {
        [ColumnName("output")]
        public float Prediction { get; set; }
    }

        public static void PredictWithMLContextSingle(string modelPath)
        {
            Console.WriteLine("\r\nRun PredictWithMLContext:");

            var mlContext = new MLContext();

            // Define the ONNX pipeline
            var pipeline = mlContext.Transforms.ApplyOnnxModel(
                modelFile: modelPath,
                outputColumnNames: new[] { "output" },
                inputColumnNames: new[] { "input" });
            // Prepare input data (replace this with your actual input data)
            var input = new ModelInput();
            for (int i = 0; i < input.Features.Length; i++)
            {
                input.Features[i] = 0.1f; // Example value, replace with your actual data
            }

            // Create the data view with a single item wrapped in a list
            var emptyDataView = mlContext.Data.LoadFromEnumerable(new[] { input });

            // Fit the pipeline
            var mlModel = pipeline.Fit(emptyDataView);

            // Load the model
            var predictionEngine = mlContext.Model.CreatePredictionEngine<ModelInput, ModelOutput>(mlModel);

            // Perform prediction
            var output = predictionEngine.Predict(input);

            Console.WriteLine("Using MLContext Prediction result:");
            Console.WriteLine($"Output: {output}");
        }

Expected behavior
Prediction works with MLContext.

Screenshots, Code, Sample Projects
Sample code in https://github.com/xqiu/ml_onnx_output_error

Additional context
Add any other context about the problem here.

[verdie-g]

Here is a self-contained reproduction

MLContext mlContext = new();

string[] inputColumns = ["A", "B", "C"];
string[] outputColumns = ["X"];

OnnxScoringEstimator estimator = mlContext.Transforms
    .ApplyOnnxModel(
        outputColumnNames: outputColumns,
        inputColumnNames: inputColumns,
        new MemoryStream(Convert.FromBase64String("CAo6aQoOCgFBCgFCEgFEIgNBZGQKDgoBQwoBRBIBRSIDQWRkCg8KAUUSAVgiB1NxdWVlemUSAmxyWgsKAUESBgoECAESAFoLCgFCEgYKBAgBEgBaCwoBQxIGCgQIARIAYgsKAVgSBgoECAESAEIECgAQFQ==")));

IDataView emptyDv = mlContext.Data.LoadFromEnumerable(Array.Empty<OnnxInput>());

OnnxTransformer transformer = estimator.Fit(emptyDv);

PredictionEngine<OnnxInput, OnnxOutput> predictionEngine = mlContext.Model.CreatePredictionEngine<OnnxInput, OnnxOutput>(transformer);

OnnxOutput output = predictionEngine.Predict(new OnnxInput { A = 1, B = 2, C = 3 });

Console.WriteLine(output.X);

public class OnnxInput
{
    [ColumnName("A")]
    public float A { get; set; }

    [ColumnName("B")]
    public float B { get; set; }

    [ColumnName("C")]
    public float C { get; set; }
}

public class OnnxOutput
{
    [ColumnName("X")]
    public float X { get; set; }
}