public void Analyse(List <IProbabilityDistribution> inputDistributions, List <IProbabilityDistribution> outputDistributions, WorkflowComponent innerWorkflow)
{
using (filer = new CSVFiler(path))
{
int NVariables = inputDistributions.Count;
int NTargets = outputDistributions.Count;
var sampler = new FASTSampler(NVariables);
int NSamples = sampler.Ns;
int Ns2 = NSamples / 2;
propagator.Propagate(inputDistributions, outputDistributions, innerWorkflow);
Matrix <double> samples = propagator.Samples;
double[] ResultMeans = new double[NTargets];
double[] ResultVariances = new double[NTargets];
double[] ResultStandardDeviations = new double[NTargets];
for (int r = 0; r < NTargets; r++) //for rth result (output)
{
//to store intermedient sum
double sum = 0;
double sumSquared = 0;
int r2 = NVariables + r;
for (int s = 0; s < Ns2; s++)
{
double sample = samples[s, r2];
sum += sample;
sumSquared += sample * sample;
}
ResultMeans[r] = sum / Ns2;
ResultVariances[r] = sumSquared / Ns2 - (ResultMeans[r]) * (ResultMeans[r]); //sobol's original approach (satelli's approach to be added)
ResultStandardDeviations[r] = Sqrt(ResultVariances[r]);
}
var A = new List <double>();
var B = new List <double>();
var Lambda = new List <double>();
double[] samplesFAST = sampler.Samples;
Matrix <double> Sensitivities = Matrix <double> .Build.Dense(NTargets, NVariables);
for (int t = 0; t < NTargets; t++)
{
ResultVariances[t] = 0;
Lambda.Clear();
//for (int j = -(NSamples - 1) / 2; j <= (NSamples - 1) / 2; j++) in "A Quantitative Model-Independent Method for Global Sensitivity Analysis of Model Output"
for (int s = 0; s < Ns2; s++) //seems -(NSamples - 1) / 2 ~ 0 are not used
{
double tempFA = 0;
double tempFB = 0;
for (int k = 0; k < NSamples; k++)
{
//tempFA += ResultMatrix[k][t] * Cos(s * samplesFAST[k]);
//tempFB += ResultMatrix[k][t] * Sin(s * samplesFAST[k]);
tempFA += samples[k, NVariables + t] * Cos(s * samplesFAST[k]);
tempFB += samples[k, NVariables + t] * Sin(s * samplesFAST[k]);
}
tempFA = tempFA / NSamples;
tempFB = tempFB / NSamples;
A.Add(tempFA);
B.Add(tempFB);
double lambda = Pow(tempFA, 2) + Pow(tempFB, 2);
Lambda.Add(lambda);
ResultVariances[t] += lambda;
}
ResultVariances[t] *= 2;
ResultStandardDeviations[t] = Sqrt(ResultVariances[t]);
ResultMeans[t] = samples.SubMatrix(0, Ns2, NVariables + t, 1).Column(0).Sum() / Ns2;
// Get Sensitivities
for (int v = 0; v < NVariables; v++)
{
Sensitivities[t, v] = 0;
for (int p = 1; p <= sampler.M; p++)
{
int temp_counter = p * sampler.Omega[v] - 1;
Sensitivities[t, v] += Lambda[temp_counter];
}
Sensitivities[t, v] *= 2 / ResultVariances[t];
}
//Write .csv file
filer.NewRow();
filer.AddToRow(t);
filer.AddToRow(outputDistributions[t].Name);
foreach (var s in Sensitivities.Row(t))
{
filer.AddToRow(s);
}
filer.WriteRow();
}
for (int t = 0; t < NTargets; t++)
{
filer.NewRow();
filer.AddToRow(t);
filer.AddToRow(outputDistributions[t].Name);
for (int v = 0; v < NVariables; v++)
{
filer.AddToRow(Sensitivities[t, v]);
}
filer.AddToRow(1 - Sensitivities.Row(t).Sum());
}
}
}
public void Propagate(List <IFASTDistribution> inputDistributions, List <ProbabilityDistributionFromSamples> outputDistributions, WorkflowComponent innerWorkflow)
{
filer = (createFile) ? new CSVFiler(path) : null;
try
{
int Ninputs = inputDistributions.Count;
int Noutputs = outputDistributions.Count;
sampler = new FASTSampler(Ninputs);
int NSamples = sampler.Ns;
Matrix <double> samples = Matrix <double> .Build.Dense(NSamples, Ninputs + Noutputs);
for (int i = 0; i < inputDistributions.Count; i++)
{
samples.SetColumn(i, inputDistributions[i].GetSamples(NSamples, sampler.GetSamplesForVariable(i)));
}
for (int s = 0; s < NSamples; s++)
{
int v = 0;
foreach (IProbabilityDistribution input in inputDistributions)
{
input.Data.Value = samples[s, v];
v++;
}
// Execute workflow
bool statusToCheck = innerWorkflow.Execute();
foreach (IProbabilityDistribution output in outputDistributions)
{
samples[s, v] = Convert.ToDouble(output.Data.Value);
v++;
}
if (createFile && statusToCheck)
{
// Execute database insert command
filer.NewRow();
//filer.AddToRow(i);
foreach (Data input in innerWorkflow.ModelDataInputs)
{
filer.AddToRow(input);
}
foreach (Data output in innerWorkflow.ModelDataInputs)
{
filer.AddToRow(output);
}
filer.WriteRow();
}
}
int o = Ninputs;
foreach (IProbabilityDistribution output in outputDistributions)
{
output.Update(samples.Column(o).AsArray());
o++;
}
Samples = samples;
}
finally
{
filer?.Dispose();
}
}
