private void ProcessPacket(OSCPacket packet)
{
foreach (OSCPacket p in packet.Data)
{
if (p.IsBundle())
{
ProcessPacket(p);
}
else
{
OSCMessage msg = (OSCMessage)p;
if ((EyesStatus)msg.Extra != EyesStatus.NONE)
{
Status = (EyesStatus)msg.Extra;
}
if (msg.Address == "/muse/elements/alpha_absolute")
{
if (!Training || (Training && alphaIgnore == 0 && keepTrainingData))
{
if (alphaSet)
{
alpha2 = (float)msg.Data[0];
alphaSet = false;
if (prevStatus == Status)
{
TrainingValue trainingValue = new TrainingValue((int)Status, 2);
trainingValue.Features[0] = alpha1;
trainingValue.Features[1] = alpha2;
AlphaTrainingValues.Add(trainingValue);
}
}
else
{
alpha1 = (float)msg.Data[0];
alphaSet = true;
}
}
else if (Training && alphaIgnore != 0)
{
alphaIgnore--;
}
prevStatus = Status;
}
else if (msg.Address == "/muse/eeg")
{
tp9.Add((float)msg.Data[0]);
af7.Add((float)msg.Data[1]);
af8.Add((float)msg.Data[2]);
tp10.Add((float)msg.Data[3]);
}
}
}
}
private void Predict()
{
TrainingValue tmp = new TrainingValue();
tmp.Features = new double[FEATURE_COUNT];
FillTrainingValue(ref tmp, fftResults);
int result = trainer.Predict(tmp);
predictedMuscleState = (MuscleState)result;
}
private void Train()
{
if (skipsRemaining > 0)
{
skipsRemaining--;
return;
}
TrainingValue value = new TrainingValue((int)currentMuscleState, FEATURE_COUNT);
FillTrainingValue(ref value, fftResults);
trainingData.Add(value);
}
public static void FillTrainingValue(ref TrainingValue value, List <Complex> fftResults)
{
float freqRange = HIGHER_FREQ - LOWER_FREQ;
float freqStep = freqRange / FEATURE_COUNT;
float lower = LOWER_FREQ;
float higher = lower + freqStep;
for (int i = 0; i < FEATURE_COUNT; i++)
{
value.Features[i] = PSD(fftResults, EMGPacket.SAMPLE_RATE, lower, higher);
lower += freqStep;
higher += freqStep;
}
}
internal static List <TrainingValue> GetTrainingValues(List <EMGPacket> packets, bool enableSkip)
{
List <TrainingValue> values = new List <TrainingValue>(EMGProcessor.FFT_SAMPLE_SIZE);
int skipsRemaining = 0;
for (int i = 0; i < packets.Count / EMGProcessor.FFT_SAMPLE_SIZE; i++)
{
if (enableSkip && skipsRemaining > 0)
{
skipsRemaining--;
continue;
}
Complex[] data = new Complex[EMGProcessor.FFT_SAMPLE_SIZE];
int start = i * EMGProcessor.FFT_SAMPLE_SIZE;
int end = start + EMGProcessor.FFT_SAMPLE_SIZE;
MuscleState startMuscleState = packets[start].muscleStateHint;
for (int j = start; j < end; j++)
{
EMGPacket packet = packets[j];
if (packet.muscleStateHint != startMuscleState)
{
skipsRemaining += EMGProcessor.SKIPS_AFTER_TRANSITION;
break;
}
data[j - start] = new Complex(EMGProcessor.ValueFromPacket(packet), 0);
}
if (enableSkip && skipsRemaining > 0)
{
continue;
}
FourierTransform.FFT(data, FourierTransform.Direction.Forward);
List <Complex> fftResults = new List <Complex>(data);
TrainingValue trainingValue = new TrainingValue((int)packets[start].muscleStateHint, EMGProcessor.FEATURE_COUNT);
EMGProcessor.FillTrainingValue(ref trainingValue, fftResults);
values.Add(trainingValue);
}
return(values);
}
private void RunFFT()
{
readingsMean.Clear();
for (int i = 0; i < af7.Count; i++)
{
readingsMean.Add(af7[i] + tp9[i] + tp10[i] + af8[i] / FEATURE_COUNT);
}
CalculateFFT(readingsMean, FFTResults);
CalculateFFT(af7, AF7FFT);
CalculateFFT(af8, AF8FFT);
CalculateFFT(tp9, TP9FFT);
CalculateFFT(tp10, TP10FFT);
if (!Training || (Training && ignore == 0))
{
TrainingValue trainingValue = new TrainingValue((int)Status, FEATURE_COUNT);
trainingValue.Features[0] = PSD(TP9FFT, FREQ_STEP);
trainingValue.Features[1] = PSD(AF7FFT, FREQ_STEP);
trainingValue.Features[2] = PSD(AF8FFT, FREQ_STEP);
trainingValue.Features[3] = PSD(TP10FFT, FREQ_STEP);
if (!Training && trainer != null && trainer.Trained)
{
Status = (EyesStatus)trainer.Predict(trainingValue);
}
if (training || keepTrainingData)
{
TrainingValues.Add(trainingValue);
}
}
else if (Training && ignore != 0)
{
ignore--;
}
}
public int Predict(TrainingValue val)
{
if (!Trained)
{
throw new Exception("Train must be called first!");
}
switch (type)
{
case ClassifierType.DecisionTree:
return(tree.Decide(val.Features));
case ClassifierType.LDA:
return(pipeline.Decide(val.Features));
case ClassifierType.SVM:
return(Convert.ToInt32(svm.Decide(val.Features)));
case ClassifierType.Bayes:
return(bayes.Decide(val.Features));
}
return(-1);
}
private string csvLineFromTrainingValue(TrainingValue val)
{
return(string.Format("{0},{1},{2}", val.Features[0], val.Features[1], (int)val.State));
}
