RhSolutions-ML/RhSolutions.ML.Lib/RhSolutionsMLBuilder.cs

using Microsoft.ML;
using Microsoft.ML.Data;

namespace RhSolutions.ML.Lib;

public class RhSolutionsMLBuilder
{
	private static string _appPath = Path.GetDirectoryName(Environment.GetCommandLineArgs()[0]) ?? ".";
	private static string _testDataPath = Path.Combine(_appPath, "..", "..", "..", "..", "TestData", "*");

	private static MLContext _mlContext = new MLContext(seed: 0);

	public static void RebuildModel()
	{
		var _trainDataView = _mlContext.Data.LoadFromTextFile<Product>(
			Path.Combine(_appPath, "..", "..", "..", "..", "TrainData", "*"), hasHeader: false);
		var pipeline = ProcessData();
		BuildAndTrainModel(_trainDataView, pipeline, out ITransformer _trainedModel);
		SaveModelAsFile(_mlContext, _trainDataView.Schema, _trainedModel);
	}

	public static MulticlassClassificationMetrics EvaluateModel()
	{
		var testDataView = _mlContext.Data.LoadFromTextFile<Product>(_testDataPath, hasHeader: false, separatorChar: ';');
		MLContext mlContext = new(seed: 0);		
		string modelPath = Path.Combine(_appPath, "..", "..", "..", "..", "Models", "model.zip");
		var trainedModel = mlContext.Model.Load(modelPath, out _);
		return _mlContext.MulticlassClassification.Evaluate(trainedModel.Transform(testDataView));
	}

	private static IEstimator<ITransformer> ProcessData()
	{
		var pipeline = _mlContext.Transforms.Conversion.MapValueToKey(inputColumnName: "Type", outputColumnName: "Label")
			.Append(_mlContext.Transforms.Text.FeaturizeText(inputColumnName: "Name", outputColumnName: "NameFeaturized"))
			.Append(_mlContext.Transforms.Concatenate("Features", "NameFeaturized"))
			.AppendCacheCheckpoint(_mlContext);
		return pipeline;
	}

	private static IEstimator<ITransformer> BuildAndTrainModel(IDataView trainingDataView, IEstimator<ITransformer> pipeline, out ITransformer trainedModel)
	{
		var trainingPipeline = pipeline.Append(_mlContext.MulticlassClassification.Trainers.SdcaMaximumEntropy("Label", "Features"))
			.Append(_mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel"));

		trainedModel = trainingPipeline.Fit(trainingDataView);
		return trainingPipeline;
	}

	private static void SaveModelAsFile(MLContext mlContext, DataViewSchema trainingDataViewSchema, ITransformer model)
	{
		string path = Path.Combine(_appPath, "..", "..", "..", "..", "Models");
		if (!Directory.Exists(path))
		{
			Directory.CreateDirectory(path);
		}
		mlContext.Model.Save(model, trainingDataViewSchema,
				Path.Combine(path, "model.zip"));
	}
}