public Data Clone()
{
Data d = new Data(this.RealInput.Clone() as double[], this.RealOutput.Clone() as double[]);
d.Input = this.Input.Clone() as double[];
d.Output = this.Output.Clone() as double[];
return d;
}
public RCRunnableConfiguration(double[][] w, double[][] B, double[][] I, double[] H0, Data[] dataSet,
int hidenNodesNumber, ERCActivationFunctionType activationFunctionType)
: base(hidenNodesNumber, activationFunctionType)
{
this.w = w;
this.B = B;
this.I = I;
this.H0 = H0;
DataSet = dataSet;
}
public DataProvider(Data[] dataSet, EExecutionType executionType, int seed, double trainSize, double validationSize, double testSize)
{
_Random = new Random(seed);
ExecutionType = executionType;
DataSet = dataSet;
DataSetLines = dataSet.Length;
DataSetColumns = dataSet[0].Input.Length + dataSet[0].Output.Length;
InputsN = dataSet[0].Input.Length;
OutputsN = dataSet[0].Output.Length;
MinValue = new double[DataSetColumns];
MaxValue = new double[DataSetColumns];
_TrainSize = trainSize;
_ValidationSize = validationSize;
_TestSize = testSize;
if ((_TrainSize + _ValidationSize + _TestSize) != 1)
throw new ArgumentException("A soma do percentual de particionamento dos dados deve ser 100%");
}
private double[] _GenerateSR(int N, int m, Data[] dataSet, double[][] predictionOutput)
{
double[] successCount = new double[m];
double[] sr = new double[m];
for (int i = 0; i < N; i++)
{
for (int j = 0; j < m; j++)
{
if (dataSet[i].RealOutput[j] == Math.Round(_Prov.DeNormalizeOutputData(predictionOutput[i][j],j)))
successCount[j]++;
}
}
for (int i = 0; i < m; i++)
sr[i] = successCount[i] * 100 / N;
return sr;
}
private double[] _GenerateEPMAForPowerPlant(int N, int m, Data[] dataSet, double[][] predicitionOutput, double powerPlantPotency)
{
double[] EPMA = new double[m];
double Error = 0;
for (int i = 0; i < N; i++)
{
for (int j = 0; j < m; j++)
{
Error = _Prov.DeNormalizeOutputData(predicitionOutput[i][j], j) - dataSet[i].RealOutput[j];
EPMA[j] += Math.Abs(Error / powerPlantPotency);
}
}
for(int i = 0 ; i < m; i++)
EPMA[i] = EPMA[i] / N * 100;
return EPMA;
}
private double[] _GenerateEPMA(int N, int m, Data[] dataSet, double[][] predictionOutput, bool evaluateLikeONS)
{
int epmaLength = evaluateLikeONS ? 7 : m;
double[] EPMA = new double[epmaLength];
double error = 0;
double divisor = 0;
for (int i = 0; i < N; i++)
{
for (int j = evaluateLikeONS ? 3 : 0, k = 0; j < (evaluateLikeONS ? 10 : m); j++, k++)
{
if(_KeepDataNormalized)
error = predictionOutput[i][j] - dataSet[i].Output[j];
else
error = _Prov.DeNormalizeOutputData(predictionOutput[i][j], j) - dataSet[i].RealOutput[j];
if(_KeepDataNormalized)
divisor = dataSet[i].Output[j] == 0 ? Double.Epsilon : dataSet[i].Output[j];
else
divisor = dataSet[i].RealOutput[j] == 0 ? Double.Epsilon : dataSet[i].RealOutput[j];
EPMA[k] += Math.Abs(error / divisor);
}
}
for (int i = 0; i < epmaLength; i++)
EPMA[i] = EPMA[i] / N * 100;
return EPMA;
}
private double[] _GenerateEMQ(int N, int m, Data[] dataSet, double[][] predictionOutput)
{
double[] EMQ = new double[m];
double error = 0;
for (int i = 0; i < N; i++)
{
for (int j = 0; j < m; j++)
{
if(_KeepDataNormalized)
error = dataSet[i].Output[j] - predictionOutput[i][j];
else
error = dataSet[i].RealOutput[j] - _Prov.DeNormalizeOutputData(predictionOutput[i][j], j);
EMQ[j] += Math.Pow(error, 2);
error = 0;
}
}
for (int i = 0; i < m; i++)
EMQ[i] = EMQ[i] / N;
return EMQ;
}
private double[] _GenerateDEV(int N, int m, Data[] dataSet, double[][] predictionOutput)
{
double[] Mean = new double[m];
double[] DEV = new double[m];
double error = 0;
for (int i = 0; i < N; i++)
{
for (int j = 0; j < m; j++)
{
if(_KeepDataNormalized)
error = dataSet[i].Output[j] - predictionOutput[i][j];
else
error = dataSet[i].RealOutput[j] - _Prov.DeNormalizeOutputData(predictionOutput[i][j], j);
Mean[j] += error;
error = 0;
}
}
for (int i = 0; i < m; i++)
Mean[i] = Mean[i] / N;
for (int i = 0; i < N; i++)
{
for (int j = 0; j < m; j++)
{
if (_KeepDataNormalized)
error = dataSet[i].Output[j] - predictionOutput[i][j];
else
error = dataSet[i].RealOutput[j] - _Prov.DeNormalizeOutputData(predictionOutput[i][j], j);
DEV[j] += Math.Pow(error - Mean[j], 2);
error = 0;
}
}
for (int i = 0; i < m; i++)
DEV[i] = DEV[i] / (N - 1);
return DEV;
}
private void _ReadDataSetFromFile(String filePath)
{
String line = null;
using (StreamReader reader = new StreamReader(filePath))
{
line = reader.ReadLine();
DataSetLines = int.Parse(line); // número de exemplos
line = reader.ReadLine();
DataSetColumns = int.Parse(line); // representa o número de entradas + número de saidas
line = reader.ReadLine();
InputsN = int.Parse(line);; // número de entradas da rede
line = reader.ReadLine();
OutputsN = int.Parse(line);; // número de saídas da rede
DataSet = new Data[DataSetLines];
for(int i = 0; i < DataSetLines; i++)
{
line = reader.ReadLine();
String[] values = line.Split(';');
DataSet[i] = new Data(new double[InputsN], new double[OutputsN]);
for (int j = 0; j < values.Length; j++)
{
double val = double.Parse(values[j]);
if (double.IsNaN(val))
throw new Exception();
if (j >= InputsN)
{
DataSet[i].RealOutput[j - InputsN] = val;
DataSet[i].Output[j - InputsN] = val;
}
else
{
DataSet[i].RealInput[j] = val;
DataSet[i].Input[j] = val;
}
}
}
}
}
public DataProvider ShrinkProvider(int inputs)
{
DataProvider shrinkProv = null;
if (inputs < InputsN)
{
Data[] set = new Data[DataSetLines];
for (int i = 0; i < DataSetLines; i++)
{
set[i] = new Data(new double[inputs], DataSet[i].RealOutput.Clone() as double[]);
for (int j = inputs - 1, k = InputsN - 1; j >= 0; j--, k--)
{
set[i].RealInput[j] = DataSet[i].RealInput[k];
}
}
shrinkProv = new DataProvider(set, ExecutionType, _Random.Next(), _TrainSize, _ValidationSize, _TestSize);
if (_LogAppliedToData)
shrinkProv.ApplyLogToData();
if (_DataNormalized)
shrinkProv.NormalizeData(_InputBoundA, _InputBoundB, _OutputBoundA, _OutputBoundB);
shrinkProv.ShuffleDataSet();
shrinkProv.SplitData();
}
else if (inputs == InputsN)
shrinkProv = this;
else
throw new ArgumentOutOfRangeException();
return shrinkProv;
}