public double learn()
{
int sizeOfTestSet = trainingSet.Count / settings.crossValidation_k;
List <Configuration>[] subsets = new List <Configuration> [settings.crossValidation_k];
for (int i = 0; i < settings.crossValidation_k; i++)
{
subsets[i] = new List <Configuration>();
}
// separate the training Set into k roughly equal parts
Random r = new Random(settings.crossValidation_k);
for (int i = 0; i < trainingSet.Count; i++)
{
int set = r.Next(settings.crossValidation_k);
subsets[set].Add(trainingSet[i]);
}
List <LearningRound> lastModels = new List <LearningRound>();
// learn models for each split
for (int i = 0; i < settings.crossValidation_k; i++)
{
List <Configuration> currTrainingSet = new List <Configuration>();
for (int j = 0; j < settings.crossValidation_k; j++)
{
if (j != i)
{
currTrainingSet.AddRange(subsets[j]);
}
}
Learning experiment = new MachineLearning.Learning.Regression.Learning(currTrainingSet, subsets[i]);
experiment.mlSettings = settings;
experiment.learn();
lastModels.Add(experiment.models[0].LearningHistory[experiment.models[0].LearningHistory.Count]);
}
double error = 0;
for (int i = 0; i < lastModels.Count; i++)
{
error += lastModels[i].validationError_relative;
}
return(error / settings.crossValidation_k);
}
/// <summary>
/// Perform 'K-Fold Cross Validation'. For further details <see cref="T:MachineLearning.Learning.Regression.KFoldCrossValidation"/>.
/// </summary>
/// <returns>The mean error rate of all k learning rounds.</returns>
public double learn()
{
int sizeOfTestSet = trainingSet.Count / settings.crossValidation_k;
// Shuffle the list with random seed k
trainingSet = Shuffle <Configuration>(trainingSet, settings.crossValidation_k);
// Split the list in multiple partitions
List <Configuration>[] subsets = SplitList <Configuration>(trainingSet, settings.crossValidation_k);
List <LearningRound> lastModels = new List <LearningRound>();
// learn models for each split
for (int i = 0; i < settings.crossValidation_k; i++)
{
List <Configuration> currTrainingSet = new List <Configuration>();
for (int j = 0; j < settings.crossValidation_k; j++)
{
if (j != i)
{
currTrainingSet.AddRange(subsets[j]);
}
}
Learning experiment = new MachineLearning.Learning.Regression.Learning(currTrainingSet, subsets[i]);
experiment.mlSettings = settings;
experiment.learn();
lastModels.Add(experiment.models[0].LearningHistory[experiment.models[0].LearningHistory.Count - 1]);
}
double error = 0;
for (int i = 0; i < lastModels.Count; i++)
{
error += lastModels[i].validationError_relative;
}
return(error / settings.crossValidation_k);
}