static void Main(string[] args)
{
BCCConsole.Write(BCCConsoleColor.Blue, false, "News Classification Trainer Started");
//FindTheBestModel();
TrainModel();
}
static void Main(string[] args)
{
Console.ForegroundColor = ConsoleColor.DarkRed;
Console.WriteLine("Masir");
BCCConsole.Write(BCCConsoleColor.DarkBlue, true, "Amir", "Asir");
"Masir".Yellow();
}
private static void TrainModelWithUnbalancedData()
{
BCCConsole.Write(BCCConsoleColor.Gray, false, "Trainer Base Is Started ...");
var mlContext = new MLContext(0);
string trainDataPath = Environment.CurrentDirectory + @"\Data\uci-news-aggregator.csv";
string trainCachePath = @"Cache\";
string unbalancedDataFile = "Data\\Unbalanced.csv";
CreateUnbalanceDataFile(trainDataPath, unbalancedDataFile);
var trainDataView = mlContext.Data.LoadFromTextFile <ModelInput>(
unbalancedDataFile,
hasHeader: true,
separatorChar: ',',
allowQuoting: true
);
var preProcessingPipeline = mlContext.Transforms
.Conversion.MapValueToKey("Label", "Category")
.Append(mlContext.Transforms.Text.FeaturizeText(inputColumnName: "Title", outputColumnName: "Features"))
.Append(mlContext.Transforms.NormalizeMinMax("Features"))
.AppendCacheCheckpoint(mlContext);
var trainer = mlContext.MulticlassClassification.Trainers
.OneVersusAll(mlContext.BinaryClassification.Trainers.AveragedPerceptron());
var trainingPipeline = preProcessingPipeline
.Append(trainer)
.Append(
mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel")
);
BCCConsole.Write(BCCConsoleColor.Cyan, false, "Cross Validation Is Starting . . .");
var cvResult = mlContext.MulticlassClassification
.CrossValidate(trainDataView, trainingPipeline);
BCCConsole.Write(BCCConsoleColor.DarkGreen, false, "\n",
"Cross Validation Result Metrics",
"-----------------------------------");
var micA = cvResult.Average(m => m.Metrics.MicroAccuracy).ToString("0.###").DarkGreen();
var macA = cvResult.Average(m => m.Metrics.MacroAccuracy).ToString("0.###").DarkGreen();
var logA = cvResult.Average(m => m.Metrics.LogLossReduction).ToString("0.###").DarkGreen();
BCCConsole.Write(BCCConsoleColor.DarkGreen, false, "-----------------------------------");
var finalModel = trainingPipeline.Fit(trainDataView);
var modelPath = "Model\\NewsClassificationModel.zip";
if (!Directory.Exists("Model\\"))
{
Directory.CreateDirectory("Model\\");
}
BCCConsole.Write(BCCConsoleColor.Yellow, false, "Saving Model . . .");
mlContext.Model.Save(finalModel, trainDataView.Schema, modelPath);
BCCConsole.Write(BCCConsoleColor.Green, false, "Saved !");
}
private static void HyperParameterExploration(MLContext mlContext
, IEstimator <ITransformer> dataPreProcessingPipeLine
, IDataView trainDataView)
{
var result = new List <(double RMSE
, double RSQ
, int iterations
, int approximationRank)>();
for (int iterations = 5; iterations < 100; iterations += 5)
{
for (int approximationRank = 50; approximationRank < 250; approximationRank += 50)
{
var option = new MatrixFactorizationTrainer.Options
{
MatrixColumnIndexColumnName = "UserIdEncoded",
MatrixRowIndexColumnName = "RestaurantNameEncoded",
LabelColumnName = "TotalRating",
NumberOfIterations = iterations,
ApproximationRank = approximationRank,
Quiet = true
};
var trainer = mlContext.Recommendation()
.Trainers.MatrixFactorization(option);
var completePipeLine = dataPreProcessingPipeLine.Append(trainer);
var cvMetrics = mlContext.Recommendation()
.CrossValidate(trainDataView, completePipeLine, labelColumnName: "TotalRating");
result.Add((
cvMetrics.Average(cv => cv.Metrics.RootMeanSquaredError),
cvMetrics.Average(cv => cv.Metrics.RSquared),
iterations,
approximationRank));
}
}
BCCConsole.Write(BCCConsoleColor.DarkGreen, false, "\n", "--- Hyper Parameter Exploration Result Metrics ---");
foreach (var res in result.OrderByDescending(r => r.RSQ))
{
BCCConsole.Write(BCCConsoleColor.DarkGreen, false, "\n",
$"RSQ => RSquared : {res.RSQ:#.000}",
$"RMSE => Root Error : {res.RMSE:#.000}",
$"I => Iterations : {res.iterations}",
$"AR => ApproximationRank : {res.approximationRank}"
);
}
BCCConsole.Write(BCCConsoleColor.DarkGreen, false, "\n", "---------------------------------------");
}
private static void FindTheBestModel()
{
BCCConsole.Write(BCCConsoleColor.DarkGreen, false, "\nFinding the Best Model Using AutoML");
var mlContext = new MLContext(0);
string trainDataPath = @"Data\uci-news-aggregator.csv";
string trainCachePath = @"Cache\";
string bestModelPath = @"Model\BestModelRun.zip";
var trainDataView = mlContext.Data.LoadFromTextFile <ModelInput>(
trainDataPath,
hasHeader: true,
separatorChar: ',',
allowQuoting: true
);
var preProcessingPipeline = mlContext.Transforms
.Conversion.MapValueToKey("Category", "Category");
var mappedInputData = preProcessingPipeline
.Fit(trainDataView).Transform(trainDataView);
var experimentSetting = new MulticlassExperimentSettings()
{
MaxExperimentTimeInSeconds = 300,
CacheBeforeTrainer = CacheBeforeTrainer.On,
OptimizingMetric = MulticlassClassificationMetric.MicroAccuracy,
CacheDirectory = new DirectoryInfo(trainCachePath)
};
var experiment = mlContext.Auto().CreateMulticlassClassificationExperiment(experimentSetting);
var experimentResult = experiment.Execute(
trainData: mappedInputData,
labelColumnName: "Category",
progressHandler: new MulticlassExperimentProgressHandler()
);
BCCConsole.Write(BCCConsoleColor.Yellow, false, "Metrics From Best Run ... ");
var metrics = experimentResult.BestRun.ValidationMetrics;
BCCConsole.Write(BCCConsoleColor.DarkGreen, false, $"Metric Micro Accuracy : {metrics.MicroAccuracy:0.##}");
BCCConsole.Write(BCCConsoleColor.Green, false, "Success !");
}
public static void CreateUnbalanceDataFile(string inputFilePath, string outputFilePath)
{
BCCConsole.Write(BCCConsoleColor.Blue, false, "\n", "Unbalance 10% Filter Is Starting . . .");
var inputFileRow = File.ReadAllLines(inputFilePath);
var outputFileRow = new List <string>();
outputFileRow.Add(inputFileRow.First());
int eSample = 0;
int bSample = 0;
int tSample = 0;
int mSample = 0;
var randomGenerator = new Random(0);
foreach (var row in inputFileRow.Skip(1))
{
if (row.Contains(",b,"))
{
if (randomGenerator.NextDouble() <= .1)
{
outputFileRow.Add(row);
bSample++;
}
}
else if (row.Contains(",e,"))
{
if (randomGenerator.NextDouble() <= .1)
{
outputFileRow.Add(row);
eSample++;
}
}
else if (row.Contains(",t,"))
{
if (randomGenerator.NextDouble() <= .1)
{
outputFileRow.Add(row);
tSample++;
}
}
else if (row.Contains(",m,"))
{
if (randomGenerator.NextDouble() <= .1)
{
outputFileRow.Add(row);
mSample++;
}
}
}
File.WriteAllLines(outputFilePath, outputFileRow);
BCCConsole.Write(BCCConsoleColor.DarkGreen, false,
"\n",
"---------------------------",
"Unbalance Training Test Result ",
$"--- Business Rank {bSample}",
$"--- Entertainment Rank {eSample}",
$"--- Technology Rank {tSample}",
$"--- Medical Rank {mSample}",
"---------------------------"
);
}
static void Main(string[] args)
{
BCCConsole.Write(BCCConsoleColor.DarkBlue, false, "Restaurant Recommender Is Started . . .");
MLContext mlContext = new MLContext(0);
var trainingDataFile = Environment.CurrentDirectory + @"\Data\TrainingFile.tsv";
DataPreparer.PreprocessData(trainingDataFile);
IDataView trainingDataView = mlContext.Data
.LoadFromTextFile <ModelInput>(trainingDataFile, hasHeader: true);
var dataPreProcessingPipeLine = mlContext.Transforms.Conversion
.MapValueToKey("UserIdEncoded", nameof(ModelInput.UserId))
.Append(mlContext.Transforms.Conversion
.MapValueToKey("RestaurantNameEncoded", nameof(ModelInput.RestaurantName)));
var options = new MatrixFactorizationTrainer.Options
{
MatrixColumnIndexColumnName = "UserIdEncoded",
MatrixRowIndexColumnName = "RestaurantNameEncoded",
LabelColumnName = "TotalRating",
NumberOfIterations = 10,
ApproximationRank = 200,
Quiet = true
};
var trainer = mlContext.Recommendation().Trainers.MatrixFactorization(options);
var trainerPipeLine = dataPreProcessingPipeLine.Append(trainer);
#region Not Using CV
BCCConsole.Write(BCCConsoleColor.DarkBlue, false, "\n", "Training Model");
var model = trainerPipeLine.Fit(trainingDataView);
////Test
//var testUserId = "U1134";
var predictionEngine = mlContext.Model
.CreatePredictionEngine <ModelInput, ModelOutput>(model);
//var alreadyRatedRestaurant = mlContext.Data
// .CreateEnumerable<ModelInput>(trainingDataView, false)
// .Where(r => r.UserId == testUserId)
// .Select(r => r.RestaurantName)
// .Distinct();
//var allRestaurantNames = trainingDataView
// .GetColumn<string>("RestaurantName")
// .Distinct().Where(r => !alreadyRatedRestaurant.Contains(r));
//var scoredRestaurant = allRestaurantNames
// .Select(rn =>
// {
// var prediction = predictionEngine.Predict(
// new ModelInput()
// {
// UserId = testUserId,
// RestaurantName = rn
// });
// return (RestaurantName: rn, PredictedScore: prediction.Score);
// });
//var top10Restaurant = scoredRestaurant
// .OrderByDescending(r => r.PredictedScore)
// .Take(10);
//BCCConsole.Write(BCCConsoleColor.DarkGreen,false,
// "\n",
// $"Top 10 Restaurant Name & Rate For User {testUserId}",
// "----------------------------------------------------");
//foreach (var top in top10Restaurant)
//{
// BCCConsole.Write(BCCConsoleColor.DarkGreen,false,$"Prediction Score [{top.PredictedScore:#.0}] | Restaurant Name [{top.RestaurantName}] ");
//}
//BCCConsole.Write(BCCConsoleColor.DarkGreen,false, "----------------------------------------------------");
#endregion
#region Using CV
//var cvMetrics = mlContext.Recommendation()
// .CrossValidate(trainingDataView, trainerPipeLine, labelColumnName: "TotalRating");
//var averageRMSE = cvMetrics.Average(cv => cv.Metrics.RootMeanSquaredError);
//var averageRSquared = cvMetrics.Average(cv => cv.Metrics.RSquared);
//BCCConsole.Write(BCCConsoleColor.DarkGreen, false,
// "\n",
// "Training Result Before Cross Validation (Metrics) ",
// "--------------------------------------------------",
// $"RMSE => Root Error : {averageRMSE:#.000}",
// $"RSQ => RSquared : {averageRSquared:#.000}",
// "--------------------------------------------------");
#endregion
Console.WriteLine("Enter Restaurant Name");
string rn = Console.ReadLine();
if (string.IsNullOrEmpty(rn))
{
"Error".Red();
Environment.Exit(-1);
}
var prediction = predictionEngine.Predict(new ModelInput()
{
UserId = "CLONED",
RestaurantName = rn
});
BCCConsole.Write(BCCConsoleColor.Green, false, "\n", $"Prediction Result Score : {prediction.Score:#.0} For Rincon Huasteco");
//HyperParameterExploration(mlContext, dataPreProcessingPipeLine, trainingDataView);
}