private static void CrossValidate()
{
var mlContext = new MLContext(0);
//Creating a small data set for faster cross validation
string createdInputFile = @"Data\preparedInput.tsv";
DataPreparer.CreatePreparedDataFile(createdInputFile, onlySaveSmallSubset: true);
//Load the data from the file into a DataView
IDataView inputDataView = mlContext.Data.LoadFromTextFile <ModelInput>(
path: createdInputFile,
hasHeader: true,
separatorChar: '\t',
allowQuoting: true
);
// Prepare input data to a form consumable by a machine learning model
var dataPipeline = mlContext
.Transforms
.Text
.FeaturizeText("Features", "Comment")
.Append(mlContext.Transforms.NormalizeMeanVariance("Features"))
.AppendCacheCheckpoint(mlContext);
// Create the training algorithms
var trainers = new IEstimator <ITransformer>[]
{
mlContext.BinaryClassification.Trainers.LbfgsLogisticRegression(),
mlContext.BinaryClassification.Trainers.SgdCalibrated()
}.Reverse();
foreach (var trainer in trainers)
{
var modelPipeline = dataPipeline.Append(trainer);
Console.WriteLine($"Cross validating with {trainer.GetType().Name}");
Console.WriteLine();
var crossValidationResults = mlContext
.BinaryClassification
.CrossValidate(inputDataView, modelPipeline, numberOfFolds: 5);
var averageAccuracy = crossValidationResults.Average(m => m.Metrics.Accuracy);
Console.WriteLine($"Cross validated average accuracy: {averageAccuracy:0.###}");
var averageF1Score = crossValidationResults.Average(m => m.Metrics.F1Score);
Console.WriteLine($"Cross validated average F1Score: {averageF1Score:0.###}");
}
}
private static ITransformer RetrainModel(string modelFile, string dataPreparationPipelineFile)
{
MLContext mlContext = new MLContext(0);
// Load pre trained model
ITransformer pretrainedModel = mlContext.Model.Load(modelFile, out _);
// Extract pretrained model parameters
var pretrainedModelParameters =
((ISingleFeaturePredictionTransformer
<CalibratedModelParametersBase <LinearBinaryModelParameters, PlattCalibrator> >)
pretrainedModel)
.Model.SubModel;
string dataFile = @"Data\preparedInput.tsv";
DataPreparer.CreatePreparedDataFile(dataFile, onlySaveSmallSubset: false);
IDataView trainingDataView = mlContext.Data.LoadFromTextFile <ModelInput>(
path: dataFile,
hasHeader: true,
separatorChar: '\t',
allowQuoting: true
);
// Load data preparation pipeline
ITransformer dataPrepPipeline = mlContext.Model.Load(dataPreparationPipelineFile, out _);
// Prepare input data to a form consumable by a machine learning model
var newData = dataPrepPipeline.Transform(trainingDataView);
// Retrain model
Console.WriteLine("Start retraining model");
var startTime = DateTime.Now;
var retrainedModel =
mlContext.BinaryClassification.Trainers.LbfgsLogisticRegression()
.Fit(newData, pretrainedModelParameters);
Console.WriteLine($"Model retraining finished in {(DateTime.Now - startTime).TotalSeconds} seconds");
return(retrainedModel);
}
static void Main(string[] args)
{
Console.WriteLine("Aggression scorer model builder started!");
var mlContext = new MLContext(0);
//Load data
var createdInputFile = @"Data\preparedInput.tsv";
DataPreparer.CreatePreparedDataFile(createdInputFile, onlySaveSmallSubset: true);
IDataView trainingDataView = mlContext.Data.LoadFromTextFile <ModelInput>(
path: createdInputFile,
hasHeader: true,
separatorChar: '\t',
allowQuoting: true
);
var inputDataSplit = mlContext.Data.TrainTestSplit(trainingDataView, testFraction: .2, seed: 0);
//Build pipeline
var inputDataPreparer = mlContext
.Transforms
.Text
.FeaturizeText("Features", "Comment")
.Append(mlContext.Transforms.NormalizeMeanVariance("Features"))
.AppendCacheCheckpoint(mlContext)
.Fit(inputDataSplit.TrainSet);
// Create a training algorithm
var trainer = mlContext
.BinaryClassification
.Trainers
.LbfgsLogisticRegression();
//Fit the model
Console.WriteLine("Start training model");
var startTime = DateTime.Now;
var transformedData = inputDataPreparer.Transform(inputDataSplit.TrainSet);
ITransformer model = trainer.Fit(transformedData);
Console.WriteLine($"Model training finished in {(DateTime.Now - startTime).TotalSeconds} seconds");
//Test the model
EvaluateModel(mlContext, model, inputDataPreparer.Transform(inputDataSplit.TestSet));
//Save the model
Console.WriteLine($"Saving the model");
if (!Directory.Exists("Model"))
{
Directory.CreateDirectory("Model");
}
var modelFile = @"Model\\AggressionScoreModel.zip";
mlContext.Model.Save(model, trainingDataView.Schema, modelFile);
Console.WriteLine("The model is saved to {0}", modelFile);
var dataPreparePipelineFile = "dataPreparePipeline.zip";
Console.WriteLine($"Saving the input data preparing pipeline");
mlContext.Model.Save(inputDataPreparer, trainingDataView.Schema, dataPreparePipelineFile);
Console.WriteLine("The pipeline is saved to {0}", dataPreparePipelineFile);
var retrainedModel = RetrainModel(modelFile, dataPreparePipelineFile);
var completeRetrainedPipeline = inputDataPreparer.Append(retrainedModel);
Console.WriteLine($"Saving the retrained model");
string retrainedModelFile = @"Model\\AggressionScoreRetrainedModel.zip";
mlContext.Model.Save(completeRetrainedPipeline, trainingDataView.Schema, retrainedModelFile);
Console.WriteLine("The model is saved to {0}", retrainedModelFile);
EvaluateModel(mlContext, completeRetrainedPipeline, inputDataSplit.TestSet);
}
static void Main(string[] args)
{
Console.WriteLine("Aggression scorer model builder started!");
var mlContext = new MLContext(0);
//Build pipeline
var inputDataPreparer = mlContext
.Transforms
.Text
.FeaturizeText("Features", "Comment")
.AppendCacheCheckpoint(mlContext);
// Create a training algorithm
var trainer = mlContext
.BinaryClassification
.Trainers
.LbfgsLogisticRegression();
var trainingPipeline = inputDataPreparer.Append(trainer);
//Load data
var createdInputFile = @"Data\preparedInput.tsv";
DataPreparer.CreatePreparedDataFile(createdInputFile, onlySaveSmallSubset: true);
IDataView trainingDataView = mlContext.Data.LoadFromTextFile <ModelInput>(
path: createdInputFile,
hasHeader: true,
separatorChar: '\t',
allowQuoting: true
);
//Fit the model
Console.WriteLine("Start training model");
var startTime = DateTime.Now;
ITransformer model = trainingPipeline.Fit(trainingDataView);
Console.WriteLine($"Model training finished in {(DateTime.Now - startTime).TotalSeconds} seconds");
//Test the model
//Save the model
Console.WriteLine($"Saving the model");
if (!Directory.Exists("Model"))
{
Directory.CreateDirectory("Model");
}
var modelFile = @"Model\\AggressionScoreModel.zip";
mlContext.Model.Save(model, trainingDataView.Schema, modelFile);
Console.WriteLine("The model is saved to {0}", modelFile);
}