/// <summary>
/// Given a vector x, returns the derivative of x, elementwise
/// </summary>
/// <param name="x">the input vector x</param>
/// <returns> elementwise application of the derivative of the relu function = x > 0 ? 1 : 0.01</returns>
public MathNet.Numerics.LinearAlgebra.Vector <double> CalculateDerivative(MathNet.Numerics.LinearAlgebra.Vector <double> x)
{
return(x.Map(s => s > 0 ? 1 : 0.01));
}
/// <summary>
/// This is the method that actually does the work.
/// </summary>
/// <param name="DA">The DA object is used to retrieve from inputs and store in outputs.</param>
protected override void SolveInstance(IGH_DataAccess DA)
{
//Input
Matrix mL = null;
DA.GetData(0, ref mL);
//size of n x n matrix
int n = mL.RowCount;
//Number of eigenvalues/vectors to extract
int eigsCount = n;
if (Params.Input[1].SourceCount > 0)
{
DA.GetData(1, ref eigsCount);
}
if (eigsCount > n)
{
eigsCount = n;
}
else if (eigsCount < 1)
{
eigsCount = 1;
}
//Run calculation toggle
bool run = false;
DA.GetData(2, ref run);
//Calculate
//Create lists to store data
List <double> eigenValuesOutput = new List <double>();
DataTree <double> eigenVectorsOutput = new DataTree <double>();
Matrix eigenVectorMatrixOutput = null;
if (run && mL != null)
{
//Create matrix for MathNet eigendecomposition
var matrix = MathNet.Numerics.LinearAlgebra.Matrix <double> .Build.Dense(n, n);
for (int i = 0; i < n; i++)
{
for (int j = 0; j < n; j++)
{
matrix[i, j] = mL[i, j];
}
}
//Eigendecomposition
var evd = matrix.Evd(MathNet.Numerics.LinearAlgebra.Symmetricity.Symmetric);
MathNet.Numerics.LinearAlgebra.Vector <Complex> eigsValComplex = evd.EigenValues;
MathNet.Numerics.LinearAlgebra.Vector <Double> eigsValReal = eigsValComplex.Map(c => c.Real);
MathNet.Numerics.LinearAlgebra.Matrix <double> eigsVec = evd.EigenVectors;
//Convert data back to GH types
//Eigenvalue list
for (int i = 0; i < eigsCount; i++)
{
eigenValuesOutput.Add(eigsValReal[i]);
}
//Eigenvectors
eigenVectorMatrixOutput = new Matrix(n, eigsCount);
for (int i = 0; i < n; i++)
{
for (int j = 0; j < eigsCount; j++)
{
double val = eigsVec[i, j];
GH_Path path = new GH_Path(j);
eigenVectorsOutput.Add(val, path);
eigenVectorMatrixOutput[i, j] = val;
}
}
}
//Output
DA.SetDataList(0, eigenValuesOutput);
DA.SetDataTree(1, eigenVectorsOutput);
DA.SetData(2, eigenVectorMatrixOutput);
}
/// <summary>
/// Given an input vector x, this function computes and returns max(x,0.01x) elementwise
/// </summary>
/// <param name="x">the inout vector x</param>
/// <returns> max(x,0.01x) </returns>
public MathNet.Numerics.LinearAlgebra.Vector <double> CalculateActivation(MathNet.Numerics.LinearAlgebra.Vector <double> x)
{
return(x.Map(s => s > 0 ? s : 0.01 * s));
}