Is it possible to dynamically add the Tsrc and Tdest in CreatePredictionEngine<>?

Question

I am currently trying to dynamically change the Tsrc and Tdest in MlContext.Model.CreatePredictionEngine<>(ITransformer); in ML.net. So instead of having a class for data structure (Tsrc) and a class for the prediction (Tdesc) in the code, could I somehow dynamically add the classes probably with reflection?

I have tried to load an object using JsonConvert.deseralize() and get the type of that object to be the Tsrc and Tdest.

            ITransformer loadedModel = mlContext.Model.Load(modelPath, out var schema);
            List<DataViewSchema.Column?> columnData = new List<DataViewSchema.Column?>();
            foreach (string col in columns)
            {
                DataViewSchema.Column? sch = schema.GetColumnOrNull(col);
                columnData.Add(sch);
            }
            object obj_data = JsonConvert.DeserializeObject(str_data_one);
            object obj_prediction = JsonConvert.DeserializeObject(str_data_two);



            //var prediction = mlContext.Model.CreatePredictionEngine<IrisData, IrisPrediction>(loadedModel).Predict();
            var prediction = mlContext.Model.CreatePredictionEngine<>(loadedModel).predict();

edit: The Tsrc and Tdest are actual classes, not instances of a class or a method. an example is in the commented out code there is IrisData and IrisPrediction, this were the code of each class:

     public class IrisData
    {
        [LoadColumn(0)]
        public float SepalLength;

        [LoadColumn(1)]
        public float SepalWidth;

        [LoadColumn(2)]
        public float PetalLength;

        [LoadColumn(3)]
        public float PetalWidth;

        [LoadColumn(4)]
        public string Label;
    }

    public class IrisPrediction
    {
        [ColumnName("PredictedLabel")]
        public string PredictedLabels;
    }

how would I build the classes above in real time of the program using reflection and adding it to the Tsrc and Tdest?

Possible duplicate of [How do I use reflection to call a generic method?](https://stackoverflow.com/questions/232535/how-do-i-use-reflection-to-call-a-generic-method) — thehennyy, May 08 '19 at 11:14
the thing is, it is not wanting a method, but a class. If you look at the commented out line of code, IrisData and IrisPrediction are not instances of a class or a method, but the actual class itself. — ParkerBidigare, May 08 '19 at 11:18
The same mechanisms works for non generic methods on generic types. First call `MakeGenericType()`, then `Invoke()` — thehennyy, May 08 '19 at 11:24
I updated the main question to explain a bit better. I don't know what I would invoke in the classes since they are data structures. What I am trying to do is create a custom class in real time with reflection to be inputted to the CreatePredictionEngine a=class a, b = class b. So when I run predict() it knows what column in the training data is what on the fly so I don't have to create multiple predict functions for different classes that holds the data structure. — ParkerBidigare, May 08 '19 at 11:30
I think I kinda understand a little on the other question you posted. I will test it out and update this question to if that solved it or not. — ParkerBidigare, May 08 '19 at 11:39
yea no luck on that. Maybe there is a better way to execute the ML.net predict method by using the createPredictionEngine a different way. — ParkerBidigare, May 08 '19 at 11:50
Yea sorry I dont know ML.net in paticular. The reflection part should be the same everywhere. — thehennyy, May 08 '19 at 12:59

score 0 · Answer 1 · answered May 08 '19 at 21:58

0

Maybe use the Transform method, rather than Predict, if you don't need the results in the strongly typed format.

answered May 08 '19 at 21:58

amy8374

1,450
3
17
26

Is it possible to dynamically add the Tsrc and Tdest in CreatePredictionEngine<>?

1 Answers1