public void PrepareAlgorithm(Experiment expData, NeuralNetwork ann)
{
if (ann == null || expData == null)
{
throw new Exception("Argument value cannot be null");
}
//preparing the variables
m_Network = ann;
m_Experiment = expData;
m_Parameters = m_Network.Parameters;
m_IterationCounter = 0;
IsAlgorthmPrepared = true;
StopIteration = false;
//report
var rp = new ProgressIndicatorEventArgs()
{
ReportType = ProgramState.Started,
LearningError = float.MaxValue,
CurrentIteration = 0,
LearnOutput = null,
};
ReportProgress(rp);
}
/// <summary>
/// Before we start GP prepare all neccessery information
/// </summary>
/// <param name="termSet"></param>
/// <param name="funSet"></param>
/// <param name="annParams"></param>
public override void PrepareAlgorithm(Experiment expData, ANNParameters annParams = null)
{
if (annParams == null || expData == null)
{
throw new Exception("Argument value cannot be null");
}
//reset iteration and network
if (m_Network == null)
{
m_IterationCounter = 0;
m_Network = new BPNeuralNetwork(annParams, expData.GetColumnInputCount_FromNormalizedValue(), expData.GetColumnOutputCount_FromNormalizedValue());
m_Network.InitializeNetwork();
}
m_Experiment = expData;
m_Parameters = annParams;
IsAlgorthmPrepared = true;
StopIteration = false;
m_expRowCount = m_Experiment.GetRowCount();
//Send report for iteration
var rp = new ProgressIndicatorEventArgs()
{
ReportType = ProgramState.Started,
LearningError = -1,
CurrentIteration = 0,
LearnOutput = null,
};
ReportProgress(rp);
}
/// <summary>
/// Set GUI based on the GPParameters
/// </summary>
/// <param name="parameters"></param>
/// <returns></returns>
public bool SetParameters(ANNParameters parameters)
{
//Activation function
if (parameters.m_ActFunction is Linear)
{
cmbActivationFuncs.SelectedIndex = 0;
}
else if (parameters.m_ActFunction is Binary)
{
cmbActivationFuncs.SelectedIndex = 1;
}
else if (parameters.m_ActFunction is Sigmoid)
{
cmbActivationFuncs.SelectedIndex = 2;
}
else if (parameters.m_ActFunction is BipolarSign)
{
cmbActivationFuncs.SelectedIndex = 3;
}
else if (parameters.m_ActFunction is TanH)
{
cmbActivationFuncs.SelectedIndex = 4;
}
//larning algoritm
if (parameters.m_LearningAlgo == LearningAlgoritm.BP)
{
cmbLearningAlgorithm.SelectedIndex = 0;
}
else if (parameters.m_LearningAlgo == LearningAlgoritm.PSO)
{
cmbLearningAlgorithm.SelectedIndex = 1;
}
else
{
cmbLearningAlgorithm.SelectedIndex = 2;
}
txtActFunParam1.Text = parameters.m_ActFuncParam1.ToString();
txtMomentum.Text = parameters.m_Momentum.ToString();
textLearningRate.Text = parameters.m_LearningRate.ToString();
textNumHiddenLayers.Text = parameters.m_NumHiddenLayers.ToString();
textNeuronsOfEachLAyer.Text = parameters.m_NeuronsInHiddenLayer.ToString();
textParticles.Text = parameters.m_PSOParameters.m_ParticlesNumber.ToString();
textIWeight.Text = parameters.m_PSOParameters.m_IWeight.ToString();
textCWeight.Text = parameters.m_PSOParameters.m_GWeight.ToString();
textSWeight.Text = parameters.m_PSOParameters.m_LWeight.ToString();
return(true);
}
public ANNSettingsPanel()
{
InitializeComponent();
LoadActivationFunsInCombo();
LoadSelectionMethodsInCOmbo();
var p = new ANNParameters();
SetParameters(p);
}
/// <summary>
///
/// </summary>
/// <param name="param"></param>
/// <param name="normalizedInputCount">number of input variables after normalization. In case of category column type we need to apply 1 og N rule.</param>
/// <param name="normalizedOutputCount">number of output variables after normalization. In case of category column type we need to apply 1 og N rule.</param>
public BCNeuralNetwork(ANNParameters param, int normalizedInputCount, int normalizedOutputCount)
: base(param, normalizedInputCount, normalizedOutputCount)
{
}
public NeuralNetwork(ANNParameters param, int inputCount, int outputCount)
{
m_Parameters = param;
m_InputCount = inputCount;
m_OutputCount = outputCount;
}
public abstract void PrepareAlgorithm(Experiment expData, ANNParameters annParams = null);
/// <summary>
/// Resets factory to initial state
/// </summary>
public void ResetFactory()
{
m_Network = null;
m_IterationCounter = 0;
m_Parameters = null;
}
public void SetANNParameters(ANNParameters annp)
{
m_Parameters = annp;
}
/// <summary>
/// Return current value of GP params
/// </summary>
/// <returns></returns>
public ANNParameters GetParameters()
{
ANNParameters parameters = new ANNParameters();
if (!double.TryParse(txtMomentum.Text, out parameters.m_Momentum))
{
MessageBox.Show("Invalid value for Momentum!");
return(null);
}
if (!double.TryParse(textLearningRate.Text, out parameters.m_LearningRate))
{
MessageBox.Show("Invalid value for Learning Rate!");
return(null);
}
if (!int.TryParse(textNumHiddenLayers.Text, out parameters.m_NumHiddenLayers))
{
MessageBox.Show("Invalid value for number of Layers!");
return(null);
}
if (!int.TryParse(textNeuronsOfEachLAyer.Text, out parameters.m_NeuronsInHiddenLayer))
{
MessageBox.Show("Invalid value for number of Neurons in hidden Layer!");
return(null);
}
//parameters for activation function
if (!double.TryParse(txtActFunParam1.Text, out parameters.m_ActFuncParam1))
{
MessageBox.Show("Invalid value for Parameter 1!");
return(null);
}
//activation function
if (cmbActivationFuncs.SelectedIndex == 0)
{
parameters.m_ActFunction = new Linear();
}
else if (cmbActivationFuncs.SelectedIndex == 1)
{
parameters.m_ActFunction = new Binary();
}
else if (cmbActivationFuncs.SelectedIndex == 2)
{
parameters.m_ActFunction = new Sigmoid(parameters.m_ActFuncParam1);
}
else if (cmbActivationFuncs.SelectedIndex == 3)
{
parameters.m_ActFunction = new BipolarSign(parameters.m_ActFuncParam1);
}
else
{
parameters.m_ActFunction = new TanH();
}
//learning algo
if (cmbLearningAlgorithm.SelectedIndex == 0)
{
parameters.m_LearningAlgo = LearningAlgoritm.BP;
}
else if (cmbLearningAlgorithm.SelectedIndex == 1)
{
parameters.m_LearningAlgo = LearningAlgoritm.PSO;
}
else
{
parameters.m_LearningAlgo = LearningAlgoritm.BP;
}
if (!int.TryParse(textParticles.Text, out parameters.m_PSOParameters.m_ParticlesNumber))
{
MessageBox.Show("Invalid value for Number of Particles!");
return(null);
}
if (!double.TryParse(textIWeight.Text, out parameters.m_PSOParameters.m_IWeight))
{
MessageBox.Show("Invalid value for Inertia Weight!");
return(null);
}
if (!double.TryParse(textCWeight.Text, out parameters.m_PSOParameters.m_LWeight))
{
MessageBox.Show("Invalid value for Cognitive Weight!");
return(null);
}
if (!double.TryParse(textParticles.Text, out parameters.m_PSOParameters.m_GWeight))
{
MessageBox.Show("Invalid value for Social Weight!");
return(null);
}
return(parameters);
}
public BPNeuralNetwork(ANNParameters param, int inputCount, int outputCount)
: base(param, inputCount, outputCount)
{
}
/// <summary>
/// Before we start solver prepare all neccessery information
/// </summary>
/// <param name="termSet"></param>
/// <param name="funSet"></param>
/// <param name="annParams"></param>
public override void PrepareAlgorithm(Experiment expData, ANNParameters annParams = null)
{
if (annParams == null || expData == null)
{
throw new Exception("Argument value cannot be null");
}
//reset iteration and network
if (m_Network == null)
{
m_IterationCounter = 0;
//depending on the type of the colum create adequate neural network
var colType = expData.GetOutputColumnType();
if (colType == ColumnDataType.Binary)//Binary Clasification
{
m_Network = new BCNeuralNetwork(annParams, expData.GetColumnInputCount_FromNormalizedValue(), expData.GetColumnOutputCount_FromNormalizedValue());
}
else//multiclass classification
{
m_Network = new MCNeuralNetwork(annParams, expData.GetColumnInputCount_FromNormalizedValue(), expData.GetColumnOutputCount_FromNormalizedValue());
}
m_Network.InitializeNetwork();
}
//
m_Experiment = expData;
PSOParameters swarm = null;
if (m_Parameters != null)
{
swarm = m_Parameters.m_PSOParameters;
}
else
{
m_psoAlgorithm = null;
swarm = annParams.m_PSOParameters;
}
//
m_Parameters = annParams;
m_Parameters.m_PSOParameters = swarm;
m_expRowCount = m_Experiment.GetRowCount();
IsAlgorthmPrepared = true;
StopIteration = false;
float newfitness = 0;
if (m_psoAlgorithm == null)
{
//initilaize swarm
m_Parameters.m_PSOParameters.m_Dimension = m_Network.GetWeightsAndBiasCout();
m_psoAlgorithm = new ParticleSwarm(m_Parameters.m_PSOParameters, CrossEntropy);
//init
newfitness = m_psoAlgorithm.InitSwarm();
}
// else
newfitness = m_psoAlgorithm.RunSwarm();
var model = CalculateModel(false);
//Send report for iteration
var rp = new ProgressIndicatorEventArgs()
{
ReportType = ProgramState.Started,
LearningError = newfitness,
CurrentIteration = 0,
LearnOutput = model,
};
ReportProgress(rp);
}