internal override void EvaluateModel(ITransformer trainedModel)
{
var predictions = trainedModel.Transform(testData);
var metrics = mlContext.MulticlassClassification.Evaluate(data: predictions, labelColumnName: "Digit", scoreColumnName: "Score");
_metrics.Measure.Gauge.SetValue(MetricsRegistry.TrainTestRate, MetricsTags.CreateMetricsTags(new string[] { "Network" }, new string[] { nameof(MNIST28Predictor) }), TrainTestDataRate);
_metrics.Measure.Gauge.SetValue(MetricsRegistry.NetworkEvaluatingResult, MetricsTags.CreateMetricsTags(new string[] { "Network", "Algorithm", "Metric", "TrainTestRate" }, new string[] { nameof(MNIST28Predictor), this.TrainingAlgo.ToString(), "MacroAccuracy", this.TrainTestDataRate.ToString() }), metrics.MacroAccuracy);
}
internal override void EvaluateModel(ITransformer trainedModel)
{
var predictions = trainedModel.Transform(testData);
var metrics = mlContext.BinaryClassification.Evaluate(data: predictions, labelColumnName: "Label", scoreColumnName: "Score");
//Logging metrics: Test rate and accuracy
_metrics.Measure.Gauge.SetValue(MetricsRegistry.TrainTestRate, MetricsTags.CreateMetricsTags(new string[] { "Network" }, new string[] { nameof(BankNotePredictor) }), TrainTestDataRate);
_metrics.Measure.Gauge.SetValue(MetricsRegistry.NetworkEvaluatingResult, MetricsTags.CreateMetricsTags(new string[] { "Network", "Algorithm", "Metric", "TrainTestRate" }, new string[] { nameof(BankNotePredictor), this.TrainingAlgorithm.ToString(), "Accuracy", this.TrainTestDataRate.ToString() }), metrics.Accuracy);
}
internal override void SaveNetwork(ITransformer trainedModel)
{
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgo.ToString(),
"Saving Network"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
mlContext.Model.Save(trainedModel, trainingData.Schema, networkPath);
}
}
internal override void EvaluateModel(ITransformer trainedModel)
{
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgo.ToString(),
"Evaluating Network"
};
IDataView predictions;
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
predictions = trainedModel.Transform(testData);
}
var metrics = mlContext.Regression.Evaluate(data: predictions, labelColumnName: "Label", scoreColumnName: "Score");
_metrics.Measure.Gauge.SetValue(MetricsRegistry.TrainTestRate, MetricsTags.CreateMetricsTags(new string[] { "Network" }, new string[] { nameof(EuroPredictor) }), this.TrainTestDataRate);
_metrics.Measure.Gauge.SetValue(MetricsRegistry.NetworkEvaluatingResult, MetricsTags.CreateMetricsTags(new string[] { "Network", "Algorithm", "Metric", "TrainTestRate" }, new string[] { nameof(EuroPredictor), this.TrainingAlgo.ToString(), "MeanAbsoluteError", this.TrainTestDataRate.ToString() }), metrics.MeanAbsoluteError);
}
internal override IDataView LoadData(MLContext context, string dataPath)
{
IDataView loadedData = null;
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgorithm.ToString(),
"Loading Data"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
loadedData = context.Data.LoadFromTextFile <BankNotesInput>(path: dataPath,
hasHeader: false,
separatorChar: ','
);
}
return(loadedData);
}
public override void ProcessNetwork()
{
MLContext mlContext = new MLContext();
IDataView dataView = LoadData(mlContext, dataPath);
TrainTestData allData = mlContext.Data.TrainTestSplit(dataView, testFraction: trainTestDataRate);
trainingData = allData.TrainSet;
testData = allData.TestSet;
var dataprocessPipeline = mlContext.Transforms.CopyColumns(outputColumnName: "Label", inputColumnName: nameof(EuroDataModel.Value))
//Normalizing data
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev1)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev2)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev3)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev4)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev5)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev6)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev7)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev8)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev9)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev10)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev11)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev12)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev13)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev14)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev15)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev16)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev17)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev18)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev19)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev20)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev21)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev22)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev23)))
.Append(mlContext.Transforms.NormalizeMeanVariance(outputColumnName: nameof(EuroDataModel.DiffPrev24)))
//Concatenating input features
.Append(mlContext.Transforms.Concatenate("Features", nameof(EuroDataModel.DiffPrev1), nameof(EuroDataModel.DiffPrev2), nameof(EuroDataModel.DiffPrev3), nameof(EuroDataModel.DiffPrev4), nameof(EuroDataModel.DiffPrev5),
nameof(EuroDataModel.DiffPrev6), nameof(EuroDataModel.DiffPrev7), nameof(EuroDataModel.DiffPrev8), nameof(EuroDataModel.DiffPrev9), nameof(EuroDataModel.DiffPrev10), nameof(EuroDataModel.DiffPrev11), nameof(EuroDataModel.DiffPrev12),
nameof(EuroDataModel.DiffPrev13), nameof(EuroDataModel.DiffPrev14), nameof(EuroDataModel.DiffPrev15), nameof(EuroDataModel.DiffPrev16), nameof(EuroDataModel.DiffPrev17), nameof(EuroDataModel.DiffPrev18), nameof(EuroDataModel.DiffPrev19),
nameof(EuroDataModel.DiffPrev20), nameof(EuroDataModel.DiffPrev21), nameof(EuroDataModel.DiffPrev22), nameof(EuroDataModel.DiffPrev23), nameof(EuroDataModel.DiffPrev24)));
//Applying training algorithm
var trainer = ApplyRegressionTrainingAlgorithm(mlContext, TrainingAlgo);
var trainingPipeline = dataprocessPipeline.Append(trainer);
//Training the model
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgo.ToString(),
"Training the model"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
trainedModel = trainingPipeline.Fit(trainingData);
}
EvaluateModel(trainedModel);
SaveNetwork(trainedModel);
TestSomePredictions();
}
public override void ProcessNetwork()
{
EstimatorChain <KeyToValueMappingTransformer> pipeline = null;
try
{
IDataView loadedData = null;
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgo.ToString(),
"Loading Data"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
loadedData = LoadData(mlContext, DataPath);
}
TrainTestData allData = mlContext.Data.TrainTestSplit(loadedData, testFraction: trainTestDataRate);
trainingData = allData.TrainSet;
testData = allData.TestSet;
}
catch (Exception e)
{
MessageBox.Show("Could not load data: " + e.Message);
}
try
{
pipeline = ConfigureNetwork();
}
catch (Exception e)
{
MessageBox.Show("Could not set pipeline: " + e.Message);
}
try
{
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgo.ToString(),
"Training the model"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
trainedModel = Train(pipeline);
}
}
catch (Exception e)
{
MessageBox.Show("Training was not successful: " + e.Message);
}
try
{
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgo.ToString(),
"Evaluating Network"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
EvaluateModel(trainedModel);
}
}
catch (Exception e)
{
MessageBox.Show("Evaluating the model was not successful: " + e.Message);
}
try
{
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgo.ToString(),
"Saving Network"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
SaveNetwork(trainedModel);
}
}
catch (Exception e)
{
MessageBox.Show("Saving the network was not successful: " + e.Message);
}
try
{
TestSomePredictions();
}
catch (Exception e)
{
MessageBox.Show("Loading and using the network was not successful: " + e.Message);
}
}
public override void ProcessNetwork()
{
EstimatorChain <ITransformer> pipeline = null;
try
{
IDataView dataView = LoadData(mlContext, dataPath);
AppendTrainingTestDataRate(dataView);
}
catch (Exception e)
{
MessageBox.Show("Could not load data: " + e.Message);
}
try
{
pipeline = ConfigureNetwork();
}
catch (Exception e)
{
MessageBox.Show("Could not set pipeline: " + e.Message);
}
try
{
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgorithm.ToString(),
"Training the model"
};
//Measuring and logging training duration
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
trainedModel = Train(pipeline);
}
}
catch (Exception e)
{
MessageBox.Show("Training was not successful: " + e.Message);
}
try
{
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgorithm.ToString(),
"Evaluating Network"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
EvaluateModel(trainedModel);
}
}
catch (Exception e)
{
MessageBox.Show("Evaluating the model was not successful: " + e.Message);
}
try
{
string[] tags = new string[] {
this.ToString(),
this.TrainingAlgorithm.ToString(),
"Saving Network"
};
using (_metrics.Measure.Timer.Time(MetricsRegistry.Timer, MetricsTags.CreateMetricsTags(tags)))
{
SaveNetwork(trainedModel);
}
}
catch (Exception e)
{
MessageBox.Show("Saving the network was not successful: " + e.Message);
}
try
{
TestSomePredictions();
}
catch (Exception e)
{
MessageBox.Show("Loading and using the network was not successful: " + e.Message);
}
}