public void Run()
{
if (_ValidationDataTable == null)
{
throw new Exception("Validation Engine (ID = " + ID.ToString() + " cannot start training because TrainingDataTable is Empty/NULL/Invalid");
}
_MCCCalculator.Reset();
for (patternIndex = 0; patternIndex < InputPatternCount; patternIndex++)
{
//Feed Farword
patternRow = _ValidationDataTable.Rows[patternIndex];
_RuntimeEngine.PatternRow = patternRow;
y = _RuntimeEngine.Run();
//Feed Backword
//OutputLayer:- dpj=(Tpj-Ypj) Ypj (1-Ypj)
t = float.Parse(patternRow[_OutputAttribute].ToString());
predictionResult = _MCCCalculator.Update(t, y);
_NeuralNetwork.RaiseTaskProgressEvent(this, 0.00f, "Validation result of pattern no. " + ((int)(patternIndex + 1)).ToString() + " is T = " + t.ToString() + " and Y = " + y.ToString() + " difference is: " + ((float)(t - y)).ToString() + " [" + predictionResult + "]");
}
}
public void Run()
{
if (TrainingDataTable == null)
{
throw new Exception("Training Engine (ID = " + ID.ToString() + " cannot start training because TrainingDataTable is Empty/NULL/Invalid");
}
_MCCCalculator.Reset();
_TSSE = 0.00f;
_RMSE = 0.00f;
for (patternIndex = 0; patternIndex < InputPatternCount; patternIndex++)
{
//Feed Farword
patternRow = _TrainingData.Rows[patternIndex];
_RuntimeEngine.PatternRow = patternRow;
y = _RuntimeEngine.Run();
//Feed Backword
//OutputLayer:- dpj=(Tpj-Ypj) Ypj (1-Ypj)
t = float.Parse(patternRow[_OutputAttribute].ToString());
_MCCCalculator.Update(t, y);
error = ((t - y) * y * (1 - y));
neuron = _NeuralNetwork._OutputLayer._ListOfNeurons[0];
_NeuronErrorDictionary[neuron.ID] = error;
//Compute TSSE and RMSE
_TSSE = _TSSE + ((t - y) * (t - y)) / 2;
_RMSE = (float)Math.Sqrt((2 * _TSSE) / InputPatternCount);
//HiddenLayer:- dpj= Ypj (1-Ypj) Sum(&pk Wkj); where &pk is the error post synaptic neuron k
totalLayers = _NeuralNetwork._ListOfLayers.Count;
for (hiddenLayerIndex = totalLayers - 2; hiddenLayerIndex > 0; hiddenLayerIndex--)
{
layer = _NeuralNetwork._ListOfLayers[hiddenLayerIndex];
totalNeurons = layer._ListOfNeurons.Count;
for (neuronIndex = 0; neuronIndex < totalNeurons; neuronIndex++)
{
neuron = layer._ListOfNeurons[neuronIndex];
y = neuron._Value;
postSynapticErrorSignal = 0;
postLayer = _NeuralNetwork._ListOfLayers[hiddenLayerIndex + 1];
for (postSynapticNeuronIndex = 0; postSynapticNeuronIndex < postLayer._ListOfNeurons.Count; postSynapticNeuronIndex++)
{
postSynapticNeuron = postLayer._ListOfNeurons[postSynapticNeuronIndex];
postSynapticNeuronError = _NeuronErrorDictionary[postSynapticNeuron.ID];
postSynapticConnection = _NeuralNetwork._SourceTargetNeuronSynapticDictionary[neuron.ID.ToString() + "-" + postSynapticNeuron.ID.ToString()];
postSynapticErrorSignal = postSynapticErrorSignal + (postSynapticNeuronError * postSynapticConnection._Weight);
}
error = y * (1 - y) * postSynapticErrorSignal;
_NeuronErrorDictionary[neuron.ID] = error;
}
}
//Change Weights
for (synapticIndex = 0; synapticIndex < _NeuralNetwork._ListOfSynapticConnections.Count; synapticIndex++)
{
synaptic = _NeuralNetwork._ListOfSynapticConnections[synapticIndex];
deltaWeight = synaptic._SourceNeuron._Value * _NeuronErrorDictionary[synaptic._TargetNeuron.ID] * _NeuralNetwork._LearningRate;
synaptic._Weight = synaptic._Weight + deltaWeight + (_Momentum * _PreviousWeightDictionary[synaptic.ID]);
_PreviousWeightDictionary[synaptic.ID] = deltaWeight;
}
//Progress report
// _NeuralNetwork.RaiseTaskProgressEvent(this ,((((float)(patternIndex + 1)) / ((float)(InputPatternCount))) * 100) ,"Training Engine Learning Pattern No: " + patternIndex.ToString() + " of " + InputPatternCount.ToString());
}
}
