/// <summary>
/// Computes the readout units.
/// </summary>
private CompositeResult[] ComputeReadoutUnits(double[] predictors, out double[] outputVector)
{
CompositeResult[] unitsResults = new CompositeResult[_readoutUnitCollection.Length];
outputVector = new double[_readoutUnitCollection.Length];
for (int unitIdx = 0; unitIdx < _readoutUnitCollection.Length; unitIdx++)
{
double[] readoutUnitInputVector = _predictorsMapper.CreateVector(ReadoutLayerCfg.ReadoutUnitsCfg.ReadoutUnitCfgCollection[unitIdx].Name, predictors);
CompositeResult unitResult = _readoutUnitCollection[unitIdx].Compute(readoutUnitInputVector);
outputVector[unitIdx] = unitResult.Result[0];
unitsResults[unitIdx] = unitResult;
}
return(unitsResults);
}
/// <summary>
/// Computes the "One Takes All" group.
/// </summary>
/// <param name="allReadoutUnitResults">The collection of all readout units composite results.</param>
/// <param name="groupResult">The composite result of the group's probabilistic cluster chain.</param>
/// <param name="outputVector">The output vector.</param>
/// <returns>An index of the winning unit within the "One Takes All" group.</returns>
public int Compute(CompositeResult[] allReadoutUnitResults, out CompositeResult groupResult, out double[] outputVector)
{
int winnerIdx;
groupResult = new CompositeResult();
outputVector = new double[MemberReadoutUnitIndexCollection.Count];
if (DecisionMethod == OneTakesAllDecisionMethod.Basic)
{
for (int i = 0; i < MemberReadoutUnitIndexCollection.Count; i++)
{
//Store rescaled member units results
outputVector[i] = Interval.IntZP1.Rescale(allReadoutUnitResults[MemberReadoutUnitIndexCollection[i]].Result[0], Interval.IntN1P1);
}
//Make the sum equal to 1
outputVector.ScaleToNewSum(1d);
groupResult.Result = new double[outputVector.Length];
outputVector.CopyTo(groupResult.Result, 0);
//Rescale output vector back to -1 and 1
for (int i = 0; i < MemberReadoutUnitIndexCollection.Count; i++)
{
outputVector[i] = Interval.IntN1P1.Rescale(outputVector[i], Interval.IntZP1);
}
winnerIdx = outputVector.MaxIdx();
}
else
{
double[] inputVector = CreateInputVector(allReadoutUnitResults);
outputVector = ProbabilisticClusterChain.Compute(inputVector, out List <Tuple <int, double[]> > memberNetOuputs);
groupResult.Result = new double[outputVector.Length];
outputVector.CopyTo(groupResult.Result, 0);
groupResult.SubResults = new List <double[]>();
foreach (Tuple <int, double[]> tuple in memberNetOuputs)
{
groupResult.SubResults.Add(tuple.Item2);
}
//Rescale output to -1 and 1
for (int i = 0; i < MemberReadoutUnitIndexCollection.Count; i++)
{
outputVector[i] = Interval.IntN1P1.Rescale(outputVector[i], Interval.IntZP1);
}
winnerIdx = outputVector.MaxIdx();
}
return(winnerIdx);
}
