/// <summary>
/// start a new trainingsession to learn the new imagedatas
/// </summary>
/// <param name="TrainingImageData">data (images) to learn</param>
public void StartTrainingSession(Dictionary <Char, List <ImageData> > TrainingImageData)
{
if (!NetworkWasInitialised)
{
CharactersToRecognize = new List <Char>(TrainingImageData.Keys);
WorkingNeuralNetwork = new NeuralNetwork(ImageSense.Num(), new int[] { ImageSense.Num(), 10, 10, CharactersToRecognize.Count });
WorkingNeuralNetwork.ShuffleBoth();
NetworkWasInitialised = true;
}
// we need random here
Random rand = new Random();
const float BAD_CHARACTER = -0.5f; // the expected output for a not matching character
const float GOOD_CHARACTER = 0.5f; // the expected output for a matching character
// precalculate the expected output
float[] ExpectedOutput = new float[CharactersToRecognize.Count];
for (int j = 0; j < CharactersToRecognize.Count; j++)
{
ExpectedOutput[j] = BAD_CHARACTER;
}
// remember last character to calculate the expected output faster
int LastCharacter = 0;
int TotalLearningData = 0;
for (int chars = 0; chars < CharactersToRecognize.Count; chars++)
{
TotalLearningData += TrainingImageData[CharactersToRecognize[chars]].Count;
}
// raw learning - data
float[][] X = new float[TotalLearningData][];
float[][] Y = new float[TotalLearningData][];
int LearningdataPointer = 0;
int NumberOfInputs = ImageSense.Num();
int NumberOfOutputs = CharactersToRecognize.Count;
for (int CurrentCharacter = 0; CurrentCharacter < CharactersToRecognize.Count; CurrentCharacter++)
{
ExpectedOutput[LastCharacter] = BAD_CHARACTER;
ExpectedOutput[CurrentCharacter] = GOOD_CHARACTER;
LastCharacter = CurrentCharacter;
if (TrainingImageData[CharactersToRecognize[CurrentCharacter]].Count != 0)
{
for (int imgs = 0; imgs < TrainingImageData[CharactersToRecognize[CurrentCharacter]].Count; imgs++)
{
float[] CurrentState = ImageSense.GetMagicMatchSticksState(TrainingImageData[CharactersToRecognize[CurrentCharacter]][imgs]);
X[LearningdataPointer] = new float[NumberOfInputs];
Y[LearningdataPointer] = new float[NumberOfOutputs];
for (int a = 0; a < NumberOfInputs; a++)
{
X[LearningdataPointer][a] = CurrentState[a];
}
for (int a = 0; a < NumberOfOutputs; a++)
{
Y[LearningdataPointer][a] = ExpectedOutput[a];
}
LearningdataPointer++;
}
}
}
WorkingNeuralNetwork.Teacher.MaximumOfEpochs = 1000;
WorkingNeuralNetwork.Learn(X, Y);
}
