c# - Using ML.NET's CreatePredictionEngine with emitted types

Question

For a project I'm working on, I need to use ML.NET's CreatePredictionEngine method with emitted types for TSrc and TDst. I'm emitting those with System.Reflection.Emit.

Here's how I'm creating my dynamic prediction engine :

dynamic dynamicPredictionEngine;
var genericPredictionMethod = mlContext.Model.GetType().GetMethod("CreatePredictionEngine", new[] { typeof(ITransformer), typeof(DataViewSchema) });
var predictionMethod = genericPredictionMethod.MakeGenericMethod(inputObject.GetType(), outputObject.GetType());
dynamicPredictionEngine = predictionMethod.Invoke(mlContext.Model, new object[] { TrainedModel, PredictionPipeline });

Here, inputObject and outputObject are instances of my emitted classes.

Then, I'm running the prediction engine like so :

var result = dynamicPredictionEngine.Predict(inputObject)

But I get the following Exception :

'The best overloaded method match for 'Microsoft.ML.PredictionEngineBase<Object0f33ea95-c496-4c57-bd2c-728bb65dd0a9,Object5623e4bd-ba82-42cf-876f-5e11a5cb8bb2>.Predict(Object0f33ea95-c496-4c57-bd2c-728bb65dd0a9)' has some invalid arguments'

The GUID you see here are the names of my emitted classes.

To debug this, I tried to manually create a class that has the same properties as the emitted one, and ran the prediction engine again, with the manually created class as TSrc and kept the emitted one as TDst. This time, it worked fine.

I looked at the debugger to compare the emitted input class with the manually created one, and they seem to match (TaxiTrip is the name of the class I created) :

Debugger screenshot

For reference, here's how my TaxiTrip class is defined

namespace VL.ML
{
    class TaxiTrip
    {
        public string VendorId { get; set; }
        public string RateCode { get; set; }
        public float PassengerCount { get; set; }
        public float TripTime { get; set; }
        public float TripDistance { get; set; }
        public string PaymentType { get; set; }
        public float FareAmount { get; set; }
    }
}

Does anyone have an idea about what's going on here?

I already stumbled upon this question on SO, but my issue looks different : as I said if I create a class manually without any attributes (see above), Predict works just fine. Plus, it seems to be able to make use of my emitted type for TDst, so why would TSrc fail?

Happy to provide more details if necessary.

Thanks in advance!

question from:https://stackoverflow.com/questions/65829086/using-ml-nets-createpredictionengine-with-emitted-types

Answer 1

Waitting for answers