aswi2021tri-musketyri-gitlab/Server/ServerApp/Predictor/NaiveBayesClassifier.cs @ f9dd116f

//
// Author: Roman Kalivoda
//

using System;
using System.Collections.Generic;
using System.Linq;
using log4net;
using Microsoft.ML;

namespace ServerApp.Predictor
{
    /// <summary>
    /// Implementation of the naive Bayes classifier in ML.NET.
    /// </summary>
    class NaiveBayesClassifier : IPredictor
    {
        private static readonly ILog _log = LogManager.GetLogger(typeof(NaiveBayesClassifier));

        /// <summary>
        /// Context of the ML.NET framework.
        /// </summary>
        private MLContext _mlContext;

        /// <summary>
        /// Model instance
        /// </summary>
        private ITransformer _trainedModel;

        private PredictionEngine<ModelInput, ModelOutput> _predictionEngine;

        IDataView _trainingDataView;

        /// <summary>
        /// Instantiates new <c>MLContext</c>.
        /// </summary>
        public NaiveBayesClassifier()
        {
            _mlContext = new MLContext();
        }

        public NaiveBayesClassifier(string filename) : this()
        {
            DataViewSchema modelSchema;
            this._trainedModel = _mlContext.Model.Load(filename, out modelSchema);
            // TODO check if the loaded model has valid input and output schema
            this._predictionEngine = _mlContext.Model.CreatePredictionEngine<ModelInput, ModelOutput>(this._trainedModel);
        }

        public void Save(string filename)
        {
            if (this._trainingDataView is null)
            {
                throw new NullReferenceException("DataView is not set.");
            }
            if( this._trainedModel is null)
            {
                throw new NullReferenceException("Trained model instance does not exist. This predictor has not been trained yet.");
            }
            if(filename is null)
            {
                throw new ArgumentNullException(nameof(filename));
            }
            this._mlContext.Model.Save(this._trainedModel, this._trainingDataView.Schema, filename);
        }

        public void Fit(IEnumerable<ModelInput> trainInput)
        {
            this._trainingDataView = _mlContext.Data.LoadFromEnumerable(trainInput);
            var pipeline = _mlContext.Transforms.Conversion.MapValueToKey(nameof(ModelInput.Label))
                .Append(_mlContext.Transforms.Conversion.ConvertType(nameof(ModelInput.Hour)))
                .Append(_mlContext.Transforms.Concatenate("Features", 
                new[] { nameof(ModelInput.Temp), nameof(ModelInput.Rain), nameof(ModelInput.Wind), nameof(ModelInput.Hour) }))
                .Append(_mlContext.Transforms.NormalizeMeanVariance("Features", useCdf:false))
                .AppendCacheCheckpoint(_mlContext)
                .Append(_mlContext.MulticlassClassification.Trainers.NaiveBayes())
                .Append(_mlContext.Transforms.Conversion.MapKeyToValue(nameof(ModelOutput.PredictedLabel)));

            var cvResults = _mlContext.MulticlassClassification.CrossValidate(this._trainingDataView, pipeline);
            _log.Debug("Cross-validated the trained model");
            this._trainedModel = cvResults.OrderByDescending(fold => fold.Metrics.MicroAccuracy).Select(fold => fold.Model).First();
            _log.Info($"Selected the model #{cvResults.OrderByDescending(fold => fold.Metrics.MicroAccuracy).Select(fold => fold.Fold).First()} as the best.");
            this._predictionEngine = _mlContext.Model.CreatePredictionEngine<ModelInput, ModelOutput>(this._trainedModel);

        }

        public string Predict(ModelInput input)
        {
            _log.Debug($"Predicting for input: {input}");
            return this._predictionEngine.Predict(input).PredictedLabel;
        }

        public void Evaluate(IEnumerable<ModelInput> modelInputs)
        {
            var testDataView = this._mlContext.Data.LoadFromEnumerable(modelInputs);
            var data = _trainedModel.Transform(testDataView);
            var testMetrics = _mlContext.MulticlassClassification.Evaluate(data);

            Console.WriteLine($"*************************************************************************************************************");
            Console.WriteLine($"*       Metrics for Multi-class Classification model - Test Data     ");
            Console.WriteLine($"*------------------------------------------------------------------------------------------------------------");
            Console.WriteLine($"*       MicroAccuracy:    {testMetrics.MicroAccuracy:0.###}");
            Console.WriteLine($"*       MacroAccuracy:    {testMetrics.MacroAccuracy:0.###}");
            Console.WriteLine($"*       LogLoss:          {testMetrics.LogLoss:#.###}");
            Console.WriteLine($"*       LogLossReduction: {testMetrics.LogLossReduction:#.###}");
            Console.WriteLine($"*       Confusion Matrix: {testMetrics.ConfusionMatrix.GetFormattedConfusionTable()}");
            Console.WriteLine($"*************************************************************************************************************");
        }
    }
}