public async Task <ModelEvaluation> EvaluateModel()
{
try
{
//change application in run mode
IconUri = "Images/runningmodel.png";
RaisePropertyChangedEvent("IsRunning");
//init empty model evaluation
var mEval = new ModelEvaluation()
{
TrainingValue = new List <ZedGraph.PointPair>(),
ValidationValue = new List <ZedGraph.PointPair>(),
ModelValueTraining = new List <ZedGraph.PointPair>(),
ModelValueValidation = new List <ZedGraph.PointPair>(),
Classes = new List <string>(),
ModelOutputDim = 1
};
var mpt = new ModelPerformance();
mpt.DatSetName = "Training set";
var mpv = new ModelPerformance();
mpv.DatSetName = "Validation set";
//check if the trained model exists
if (string.IsNullOrEmpty(TrainingParameters.LastBestModel) || string.IsNullOrEmpty(TrainingParameters.LastBestModel.Trim(' ')))
{
await Application.Current.Dispatcher.BeginInvoke(
DispatcherPriority.Background,
new Action(
() => MainWindow.SetCursor(false)
));
return(mEval);
}
//get model full path
var modelMLPath = Project.GetMLConfigPath(Settings, Name);
//check if file exists
var fi = new FileInfo(modelMLPath);
if (!fi.Exists)
{
return(mEval);
}
//evaluate model against training data
var task1 = await Task.Run(() => Project.EvaluateMLConfig(modelMLPath, DataSetType.Training, EvaluationType.ResultExtended, ProcessDevice.Default));
var resultTrain = task1;
//evaluate model against validation data
var task2 = await Task.Run(() => Project.EvaluateMLConfig(modelMLPath, DataSetType.Validation, EvaluationType.ResultExtended, ProcessDevice.Default));
var resultValidation = task2;
if (resultTrain.Actual == null && resultTrain.Actual.Count <= 0)
{
return(mEval);
}
////prepare evaluation result
for (int i = 0; i < resultTrain.Actual.Count(); i++)
{
mEval.TrainingValue.Add(new PointPair(i + 1, resultTrain.Actual[i]));
}
for (int i = 0; i < resultTrain.Predicted.Count(); i++)
{
mEval.ModelValueTraining.Add(new PointPair(i + 1, resultTrain.Predicted[i]));
}
////no validation set defined
if (resultValidation.Actual != null && resultValidation.Actual.Count > 0)
{
for (int i = 0; i < resultValidation.Actual.Count(); i++)
{
mEval.ValidationValue.Add(new PointPair(i + 1, resultValidation.Actual[i]));
}
for (int i = 0; i < resultValidation.Predicted.Count(); i++)
{
mEval.ModelValueValidation.Add(new PointPair(i + 1, resultValidation.Predicted[i]));
}
}
////
mEval.Classes = resultTrain.OutputClasses;
mEval.ModelOutputDim = resultTrain.OutputClasses == null ? 1 : resultTrain.OutputClasses.Count;
//training performance result
mpt = MLEvaluator.CalculatePerformance(resultTrain, "Training set");
//validation performance result
mpv = MLEvaluator.CalculatePerformance(resultValidation, "Validation set");
mEval.TrainPerformance = mpt;
if (mEval.Classes != null)
{
mEval.TrainPerformance.Classes = mEval.Classes.ToArray();
}
mEval.ValidationPerformance = mpv;
if (mEval.Classes != null)
{
mEval.ValidationPerformance.Classes = mEval.Classes.ToArray();
}
ModelEvaluation = mEval;
return(mEval);
}
catch (Exception)
{
throw;
}
finally
{
//change application in normal mode
IconUri = "Images/model.png";
RaisePropertyChangedEvent("IsRunning");
}
}
public async Task <ModelEvaluation> EvaluateModel()
{
try
{
//change application in run mode
IconUri = "Images/runningmodel.png";
RaisePropertyChangedEvent("IsRunning");
//init empty model evaluation
var mEval = new ModelEvaluation()
{
TrainingValue = new List <ZedGraph.PointPair>(),
ValidationValue = new List <ZedGraph.PointPair>(),
ModelValueTraining = new List <ZedGraph.PointPair>(),
ModelValueValidation = new List <ZedGraph.PointPair>(),
Classes = new List <string>(),
ModelOutputDim = 1
};
var mpt = new ModelPerformance();
mpt.DatSetName = "Training set";
var mpv = new ModelPerformance();
mpv.DatSetName = "Validation set";
//check if the trained model exists
if (string.IsNullOrEmpty(TrainingParameters.LastBestModel) || string.IsNullOrEmpty(TrainingParameters.LastBestModel.Trim(' ')))
{
return(mEval);
}
//save model before evaluation since there is a data must be stored into model file.
Save();
//get model full path
var modelMLPath = Project.GetMLConfigPath(Settings, Name);
//check if file exists
var fi = new FileInfo(modelMLPath);
if (!fi.Exists)
{
return(mEval);
}
//evaluate model against training data
var task1 = await Task.Run(() => Project.EvaluateModel(modelMLPath, DataProcessing.Core.DataSetType.Training, EvaluationType.ResultyExtended, ProcessDevice.Default));
var resultTrain = task1;
//evaluate model against validation data
var task2 = await Task.Run(() => Project.EvaluateModel(modelMLPath, DataProcessing.Core.DataSetType.Validation, EvaluationType.ResultyExtended, ProcessDevice.Default));
var resultValidation = task2;
if (resultTrain.Actual == null && resultTrain.Actual.Count <= 0)
{
return(mEval);
}
////prepare evaluation result
for (int i = 0; i < resultTrain.Actual.Count(); i++)
{
mEval.TrainingValue.Add(new PointPair(i + 1, resultTrain.Actual[i]));
}
for (int i = 0; i < resultTrain.Predicted.Count(); i++)
{
mEval.ModelValueTraining.Add(new PointPair(i + 1, resultTrain.Predicted[i]));
}
////no validation set defined
if (resultValidation.Actual != null && resultValidation.Actual.Count > 0)
{
for (int i = 0; i < resultValidation.Actual.Count(); i++)
{
mEval.ValidationValue.Add(new PointPair(i + 1, resultValidation.Actual[i]));
}
for (int i = 0; i < resultValidation.Predicted.Count(); i++)
{
mEval.ModelValueValidation.Add(new PointPair(i + 1, resultValidation.Predicted[i]));
}
}
////
mEval.Classes = resultTrain.OutputClasses;
mEval.ModelOutputDim = resultTrain.OutputClasses == null ? 1 : resultTrain.OutputClasses.Count;
if (mEval.ModelOutputDim == 1)
{
//Training data set
var actTData = mEval.TrainingValue.Select(x => x.Y).ToArray();
var preTData = mEval.ModelValueTraining.Select(x => x.Y).ToArray();
mpt.SE = (float)actTData.SE(preTData);
mpt.RMSE = (float)actTData.RMSE(preTData);
mpt.NSE = (float)actTData.NSE(preTData);
mpt.PB = (float)actTData.PBIAS(preTData);
mpt.CORR = (float)actTData.R(preTData);
mpt.DETC = (float)actTData.R2(preTData);
//validation data set
var actVData = mEval.ValidationValue.Select(x => x.Y).ToArray();
var preVData = mEval.ModelValueValidation.Select(x => x.Y).ToArray();
if (actVData != null && actVData.Length > 0)
{
mpv.SE = (float)actVData.SE(preVData);
mpv.RMSE = (float)actVData.RMSE(preVData);
mpv.NSE = (float)actVData.NSE(preVData);
mpv.PB = (float)actVData.PBIAS(preVData);
mpv.CORR = (float)actVData.R(preVData);
mpv.DETC = (float)actVData.R2(preVData);
}
}
else if (mEval.ModelOutputDim > 1)
{
var retVal = CalculatePerformance(resultTrain.ActualEx, resultTrain.PredictedEx, null, null);
retVal.Add("Classes", mEval.Classes.ToList <object>());
mpt.PerformanceData = retVal;
//in case validation set is defined
if (resultValidation.Actual != null && resultValidation.Actual.Count > 0)
{
var retValV = CalculatePerformance(resultValidation.ActualEx, resultValidation.PredictedEx, null, null);
retValV.Add("Classes", mEval.Classes.ToList <object>());
mpv.PerformanceData = retValV;
}
}
mEval.TrainPerformance = mpt;
if (mEval.Classes != null)
{
mEval.TrainPerformance.Classes = mEval.Classes.ToArray();
}
mEval.ValidationPerformance = mpv;
if (mEval.Classes != null)
{
mEval.ValidationPerformance.Classes = mEval.Classes.ToArray();
}
ModelEvaluation = mEval;
return(mEval);
}
catch (Exception)
{
throw;
}
finally
{
//change application in normal mode
IconUri = "Images/model.png";
RaisePropertyChangedEvent("IsRunning");
}
}
