/// <summary>
/// Converts samples and labels to "TrainingSet" instances.
/// If required, applies feature scaling and mean normalization.
/// </summary>
internal void Run(IEnumerable samples, IEnumerable labels, InputSplitRatio splitRatio)
{
int samplesCount;
_features = GetFeatures(samples, labels, out samplesCount);
var samplesMatrix = GetSamplesMatrix(samples, samplesCount, out _mean, out _deviation);
var labelsVector = GetLabelsVector(labels, samplesCount);
if (_scaleAndNormalize) {
int sample = 0;
while (sample < samplesCount) {
for (int feature = 0; feature < _features.Length; feature++) {
samplesMatrix[sample, feature] = (samplesMatrix[sample, feature] - _mean[feature]) / _deviation[feature];
}
sample++;
}
}
switch (splitRatio) {
case InputSplitRatio.No:
TrainingSet = new Input(samplesMatrix, labelsVector, 0, samplesCount);
break;
case InputSplitRatio.Default:
var trainingSetSize = (int) Math.Ceiling(((double) 2/3)*samplesCount);
var crossValidationTestSize = (int)Math.Ceiling(((double)1 / 3) * samplesCount / 2);
var testSetSize = samplesCount - (trainingSetSize + crossValidationTestSize);
TrainingSet = new Input(samplesMatrix, labelsVector, 0, trainingSetSize);
CrossValidationSet = new Input(samplesMatrix, labelsVector, trainingSetSize, crossValidationTestSize);
TestSet = new Input(samplesMatrix, labelsVector, (trainingSetSize + crossValidationTestSize), testSetSize);
break;
default:
throw new NMachineException("Unexpected split type: " + splitRatio);
}
}
internal Settings()
{
LearningRate = 0.01;
ConvergenceDelta = 0.000001;
MaxIterations = 1500;
ScaleAndNormalize = true;
InputSplitRatio = InputSplitRatio.Default;
}
/// <summary>
/// Converts samples and labels to "TrainingSet" instances.
/// If required, applies feature scaling and mean normalization.
/// </summary>
internal void Run(IEnumerable samples, IEnumerable labels, InputSplitRatio splitRatio)
{
int samplesCount;
_features = GetFeatures(samples, labels, out samplesCount);
var samplesMatrix = GetSamplesMatrix(samples, samplesCount, out _mean, out _deviation);
var labelsVector = GetLabelsVector(labels, samplesCount);
if (_scaleAndNormalize)
{
int sample = 0;
while (sample < samplesCount)
{
for (int feature = 0; feature < _features.Length; feature++)
{
samplesMatrix[sample, feature] = (samplesMatrix[sample, feature] - _mean[feature]) / _deviation[feature];
}
sample++;
}
}
switch (splitRatio)
{
case InputSplitRatio.No:
TrainingSet = new Input(samplesMatrix, labelsVector, 0, samplesCount);
break;
case InputSplitRatio.Default:
var trainingSetSize = (int)Math.Ceiling(((double)2 / 3) * samplesCount);
var crossValidationTestSize = (int)Math.Ceiling(((double)1 / 3) * samplesCount / 2);
var testSetSize = samplesCount - (trainingSetSize + crossValidationTestSize);
TrainingSet = new Input(samplesMatrix, labelsVector, 0, trainingSetSize);
CrossValidationSet = new Input(samplesMatrix, labelsVector, trainingSetSize, crossValidationTestSize);
TestSet = new Input(samplesMatrix, labelsVector, (trainingSetSize + crossValidationTestSize), testSetSize);
break;
default:
throw new NMachineException("Unexpected split type: " + splitRatio);
}
}
