How to predict integer values using ML.NET?

Question

I'm looking at a the cs file here: https://www.microsoft.com/net/learn/apps/machine-learning-and-ai/ml-dotnet/get-started/windows and everything works well.

Now I'd like to improve the example: I'd like to predict a number-only data set and not a number-string dataset, for example predict the ouput of a seven segments display.

Here is my super easy dataset, the last column is the int number that I want to predict:

1,0,1,1,1,1,1,0
0,0,0,0,0,1,1,1
1,1,1,0,1,1,0,2
1,1,1,0,0,1,1,3
0,1,0,1,0,1,1,4
1,1,1,1,0,0,1,5
1,1,1,1,1,0,1,6
1,0,0,0,0,1,1,7
1,1,1,1,1,1,1,8
1,1,1,1,0,1,1,9

And here is my test code:

public class Digit
{
    [Column("0")] public float Up;

    [Column("1")] public float Middle;

    [Column("2")] public float Bottom;

    [Column("3")] public float UpLeft;
    [Column("4")] public float BottomLeft;
    [Column("5")] public float TopRight;
    [Column("6")] public float BottomRight;

    [Column("7")] [ColumnName("DigitValue")]
    public float DigitValue;
}

public class DigitPrediction
{
    [ColumnName("PredictedDigits")] public float PredictedDigits;
}

public PredictDigit()
{
    var pipeline = new LearningPipeline();
    var dataPath = Path.Combine("Segmenti", "segments.txt");
    pipeline.Add(new TextLoader<Digit>(dataPath, false, ","));
    pipeline.Add(new ColumnConcatenator("Label", "DigitValue"));
    pipeline.Add(new ColumnConcatenator("Features", "Up", "Middle", "Bottom", "UpLeft", "BottomLeft", "TopRight", "BottomRight"));
    pipeline.Add(new StochasticDualCoordinateAscentClassifier());
    var model = pipeline.Train<Digit, DigitPrediction>();
    var prediction = model.Predict(new Digit
    {
        Up = 1,
        Middle = 1,
        Bottom = 1,
        UpLeft = 1,
        BottomLeft = 1,
        TopRight = 1,
        BottomRight = 1,
    });

    Console.WriteLine($"Predicted digit is: {prediction.PredictedDigits}");
    Console.ReadLine();
}

As you can see it is very similar to the example provided except the last column ("Label") handling beacause I need to predict a number and not a string. I try with:

pipeline.Add(new ColumnConcatenator("Label", "DigitValue"));

but it does not work, exception:

Training label column 'Label' type is not valid for multi-class: Vec<R4, 1>. Type must be R4 or R8.

I'm sure I miss something but actually I cannot find anything on internet that can help me solve this problem.

UPDATE

I found that the dataset have to have a Label column like this:

[Column("7")] [ColumnName("Label")] public float Label;

and the DigitPrediction a Score column like:

public class DigitPrediction
{
    [ColumnName("Score")] public float[] Score;
}

Now the system "works" and I got as prediction.Score a Single[] value where the index associated with the higher value is the predicted value. Is it the right approach?

UPDATE 2 - Complete code example

Following the answer and other suggestions I got the right result, if you need it you can find complete code here.

Answer 1

Looks like you would need to add this field to your class DigitPrediction:

public class DigitPrediction
{
    [ColumnName("PredicatedLabel")]
    public uuint ExpectedDigit; // <-- This is the predicted value

    [ColumnName("Score")]
    public float[] Score; // <-- This is the probability that the predicted value is the right classification
}

And I think you would need to change the line where it writes the result to:

    Console.WriteLine($"Predicted digit is: {prediction.ExpectedDigit}");

One more thing, it looks like there is a bug in the API where the expected digit will be off by one but if you shift it by adding +1 to the predicted value it will be the correct value. I expect them to fix this in the future, there is an issue for it: (https://github.com/dotnet/machinelearning/issues/235)

Answer 2

You could also use try swapping ColumnConcatenator with ColumnCopier, in the Pipeline, for the Label column.

pipeline.Add(new ColumnCopier ("Label", "DigitValue"));

That will indicate the pipeline which column is the Label, but the output of ColumnCopier will not be a vector, unlike the output of ColumnConcatenator.

And could similarly add a Score column as well.