public float[] Evaluate(ITransformer transformer, DateTime dateTime)
{
if (_evaluation.ContainsKey(dateTime))
{
float[] score = new[] { _evaluation[dateTime] };
return(score);
}
Dictionary <string, float> feature = new Dictionary <string, float>();
List <Dictionary <string, float> > features = new List <Dictionary <string, float> >();
var featuresDictionnary = GetFeatures(dateTime);
if (featuresDictionnary == null)
{
return(null);
}
foreach (var item in featuresDictionnary)
{
feature.Add(item.Key, item.Value);
}
features.Add(feature);
IDataView data = new FloatsDataView(features);
MLContext mlContext = new MLContext();
IDataView result = transformer.Transform(data);
float[] scoreColumn = result.GetColumn <float>("Score").ToArray();
return(scoreColumn);
}
internal void InitializeEvaluation(ITransformer transformer)
{
_evaluation = new Dictionary <DateTime, float>();
DateTime from = _featurables.Min(c => c.Begin);
//to = _featurables.Max(c => c.End);
Period period = Period.S;
int periodCount = 5;
List <Dictionary <string, float> > features = new List <Dictionary <string, float> >();
for (DateTime date = from; date < DateTime.Now; date = date.AddPeriod(period, periodCount))
{
if (_featurables.All(c => c.HasFeatures(date)))
{
Dictionary <string, float> feature = new Dictionary <string, float>();
var featuresDictionnary = GetFeatures(date);
if (featuresDictionnary == null)
{
continue;
}
foreach (var item in featuresDictionnary)
{
feature.Add(item.Key, item.Value);
}
feature["Label"] = 0;
features.Add(feature);
_evaluation[date] = 0;
}
}
IDataView data = new FloatsDataView(features);
IDataView result = transformer.Transform(data);
float[] scoreColumn = result.GetColumn <float>("Score").ToArray();
int iterator = 0;
for (DateTime date = from; date < DateTime.Now; date = date.AddPeriod(period, periodCount))
{
if (_evaluation.ContainsKey(date))
{
_evaluation[date] = scoreColumn[iterator++];
}
}
}
public ITransformer Fit(IEstimator <ITransformer> estimator, DateTime to)
{
Console.WriteLine($"Running {estimator.GetType().Name}");
MLContext mlContext = new MLContext();
DateTime from = _featurables.Min(c => c.Begin);
//to = _featurables.Max(c => c.End);
Period period = Period.S;
int periodCount = 5;
List <Dictionary <string, float> > features = new List <Dictionary <string, float> >();
for (DateTime date = from; date < to; date = date.AddPeriod(period, periodCount))
{
if (_featurables.All(c => c.HasFeatures(date)) && _pipExpectation._futurPip.ContainsKey(date))
{
Dictionary <string, float> feature = new Dictionary <string, float>();
var featuresDictionnary = GetFeatures(date);
if (featuresDictionnary == null)
{
continue;
}
foreach (var item in featuresDictionnary)
{
feature.Add(item.Key, item.Value);
}
feature["Label"] = decimal.ToSingle(_pipExpectation._futurPip[date].Long);
features.Add(feature);
}
}
IDataView data = new FloatsDataView(features);
using (FileStream stream = new FileStream("data.tsv", FileMode.Create))
mlContext.Data.SaveAsText(data, stream);
string[] featureNames = features.First().Keys.Where(c => c != "Label").ToArray();
var pipeline = mlContext.Transforms.Concatenate("Features", featureNames)
.Append(mlContext.Transforms.NormalizeMeanVariance("Features"))
.Append(mlContext.Transforms.NormalizeMinMax("Features"))
.AppendCacheCheckpoint(mlContext);
//.Append(mlContext.Regression.Trainers.OnlineGradientDescent(numberOfIterations:10));
// Create data prep transformer
ITransformer transformData = pipeline.Fit(data);
IDataView transformedData = transformData.Transform(data);
//IEstimator<ITransformer> estimator = mlContext.Regression.Trainers.OnlineGradientDescent(numberOfIterations:10);
var cvResults = mlContext.Regression.CrossValidate(transformedData, estimator, numberOfFolds: 5);
// Apply transforms to training data
var models = cvResults.OrderByDescending(fold => fold.Metrics.RSquared).ToArray();
// Get Top Model
var topModel = models[0];
Console.WriteLine($"\tR2: { topModel.Metrics.RSquared}");
Console.WriteLine($"\tLossfunction: { topModel.Metrics.LossFunction}");
Console.WriteLine($"\tMeanAbsoluteError: { topModel.Metrics.MeanAbsoluteError}");
//RegressionMetrics trainedModelMetrics = mlContext.Regression.Evaluate(testDataPredictions);
//double rSquared = trainedModelMetrics.RSquared;
return(transformData.Append(topModel.Model));
}
