public static VectorC operator *(double d, VectorC v)
{
VectorC result = new VectorC(v.size);
for (int i = 0; i < v.size; i++)
{
result[i] = d * v[i];
}
return result;
}
public static VectorC operator *(VectorC v, double d)
{
VectorC result = new VectorC(v.size);
for (int i = 0; i < v.size; i++)
{
result[i] = v[i] * d;
}
return result;
}
public static MatrixC operator *(MatrixC m1, MatrixC m2)
{
if (m1.GetCols()!=m2.GetRows())
{
throw new ArgumentOutOfRangeException(
"Columns", m1, "The numbers of columns of the first matrix must be equal to" +
" the number of rows of the second matrix!");
}
MatrixC result = new MatrixC(m1.GetRows(), m2.GetCols());
VectorC v1 = new VectorC(m1.GetCols());
VectorC v2 = new VectorC(m2.GetRows());
for (int i = 0; i < m1.GetRows(); i++)
{
v1 = m1.GetRowVector(i);
for (int j = 0; j < m2.GetCols(); j++)
{
v2 = m2.GetColVector(j);
result[i, j] = VectorC.DotProduct(v1, v2);
}
}
return result;
}
/// <summary>
/// Implementation Convex optimization using ADMM( Stephen Boyd, Stanford) It is essentially least squares fitting with non-negativity constraints
/// An ANOVA computation is added below for significance testing although this implementation is NOT VALIDATED
/// The matrix is any reference matrix (database) and y is the experimental data. The result is a vector of coefficients for the model
/// </summary>
/// <param name="matrix"></param>
/// <param name="y"></param>
/// <returns></returns>
public VectorC MatchLibraryMatrix(MatrixC matrix, VectorC y)
{
#region ADMM (Alternating direction method of multipliers)
//regularized matrix version of least squares
//x = (A'*A)*A'*b
//rho, the wiggle factor
double rho = .001;
//Ax = b where x is the prediction, b is the experimental and A is the reference database matrix
// A = x^t*x
var A = matrix.Transpose() * matrix;
//b = xt*y
MatrixC y1 = new MatrixC(matrix.GetRows(), 1);
y1.ReplaceCol(y, 0);
var temp1 = matrix.Transpose() * y1;
VectorC b = new VectorC(temp1.GetRows());
for(int i = 0; i < temp1.GetRows();i++)
{
b[i] = temp1[i,0];
}
//create identity matrix
var I = A.Identity();
//A' = (A + rho*I)
var Aprime = MatrixC.Inverse(A + (rho * I));
double[] components = new double[b.GetSize()];
//add zeros to vectors
for (int i = 0; i < b.GetSize(); i++)
{
components[i] = 0;
}
VectorC u = new VectorC(components); //accululated error in the negative direction
VectorC z = new VectorC(components); //non-negative version of x
VectorC x = new VectorC(components); //regularized least squares
u = u.Clone();
z = z.Clone();
x = x.Clone();
double sumSquares = 0;
for (int i = 0; i < 1500; i++)
{
x = MatrixC.Transform(Aprime, (b + rho * (z - u)));
//for each element in x + u
//z = (x + u);
for (int j = 0; j < x.GetSize(); j++)
{
var temp = x[j] + u[j];
if (temp > 0)
{
z[j] = temp;
}
else
{
z[j] = 0;
}
}
if (i == 1499)
{
int test11 = 0;
}
//for each element in x - z
// u = u + (x - z);
for (int k = 0; k < u.GetSize(); k++)
{
u[k] = u[k] + (x[k] - z[k]);
}
//calculate RMS error
for (int ii = 0; ii < x.GetSize(); ii++)
{
sumSquares += (x[ii] - z[ii])*(x[ii] - z[ii]);
}
var RMS = Math.Sqrt(sumSquares / x.GetSize());
if (RMS < .001)
{
break;
}
}
#endregion
#region ANOVA
//Matrix yMatrix = Matrix.Create(y);
//Matrix database = (m.Transpose()*m).GetInverse()*m.Transpose();
////ADMM
//var predicted = m*z;
////Hat matrix. The hat matrix is used to reduce overfitting by finding outliers
////inv(x'*x)*x'*x'
//var hatMatrix = (m*(m.Transpose()*m).GetInverse())*m.Transpose();
////J matrix square matrix of ones
//Matrix JMatrix = Matrix.Create(hatMatrix.ColumnCount, hatMatrix.RowCount);
//for (int i = 0; i < JMatrix.ColumnCount; i++)
//{
// for (int j = 0; j < JMatrix.RowCount; j++)
// {
// JMatrix[i, j] = 1;
// }
//}
////SSr (2000 is the observations or m/z in our case)
////y'*[H - (1/n)J]*y
//var SSr = Y.Transpose()*((hatMatrix - (1/2000)*JMatrix))*Y;
////Mean Square Regression
//var MSR = SSr/matrix.GetLongLength(1);
////Identity matrix
//// The following constructs a nxn identity matrix:
//DenseMatrix identityMatrix = DenseMatrix.GetIdentity(2001);
////SSE
////y'*[I-H]*y
//var SSe = Y.Transpose()*(identityMatrix - hatMatrix)*Y;
//double MSquareRegression = MSR[0, 0];
////Mean Square Error
//var MSE = SSe/(2000 - (matrix.GetLongLength(1) + 1));
//double MSquareError = MSE[0, 0];
////F-statistic
////The fstat conducts a hypothesis test. It simply says, is there at least one coefficient that
////id different than zero?
//var f = MSquareRegression/MSquareError;
////get covariance matrix
//var C = MSquareError*(m.Transpose()*m).GetInverse();
////get diangonal
//var d = C.GetDiagonal();
////conduct t-test for all coefficients (vectorResult for OLS)
//for (int i = 0; i < z.Length; i++)
//{
// double tTest = z[i]/Math.Sqrt(d[i]);
//}
////populate correlation scores (vectorResult for OLS)
//for (int i = 0; i < z.Length; i++)
//{
//}
#endregion
return z;
}
public static VectorC operator +(VectorC v1, VectorC v2)
{
VectorC result = new VectorC(v1.size);
for (int i = 0; i < v1.size; i++)
{
result[i] = v1[i] + v2[i];
}
return result;
}
public VectorC GetUnitVector()
{
VectorC result = new VectorC(vector);
result.Normalize();
return result;
}
public bool Equals(VectorC v)
{
return vector == v.vector;
}
public VectorC Clone()
{
// returns a deep copy of the vector
VectorC v = new VectorC(vector);
v.vector = (double[])vector.Clone();
return v;
}
public static VectorC TriVectorProduct(VectorC v1, VectorC v2, VectorC v3)
{
if (v1.size != 3)
{
throw new ArgumentOutOfRangeException(
"v1", v1, "Vector v1 must be 3 dimensional!");
}
if (v1.size != 3)
{
throw new ArgumentOutOfRangeException(
"v2", v2, "Vector v2 must be 3 dimensional!");
}
if (v1.size != 3)
{
throw new ArgumentOutOfRangeException(
"v3", v3, "Vector v3 must be 3 dimensional!");
}
return v2 * VectorC.DotProduct(v1, v3) - v3 * VectorC.DotProduct(v1, v2);
}
public static double TriScalarProduct(VectorC v1, VectorC v2, VectorC v3)
{
if (v1.size != 3)
{
throw new ArgumentOutOfRangeException(
"v1", v1, "Vector v1 must be 3 dimensional!");
}
if (v1.size != 3)
{
throw new ArgumentOutOfRangeException(
"v2", v2, "Vector v2 must be 3 dimensional!");
}
if (v1.size != 3)
{
throw new ArgumentOutOfRangeException(
"v3", v3, "Vector v3 must be 3 dimensional!");
}
double result = v1[0] * (v2[1] * v3[2] - v2[2] * v3[1]) +
v1[1] * (v2[2] * v3[0] - v2[0] * v3[2]) +
v1[2] * (v2[0] * v3[1] - v2[1] * v3[0]);
return result;
}
public static double DotProduct(VectorC v1, VectorC v2)
{
double result = 0.0;
for (int i = 0; i < v1.size; i++)
{
result += v1[i] * v2[i];
}
return result;
}
public static VectorC CrossProduct(VectorC v1, VectorC v2)
{
if (v1.size != 3)
{
throw new ArgumentOutOfRangeException(
"v1", v1, "Vector v1 must be 3 dimensional!");
}
if (v2.size != 3)
{
throw new ArgumentOutOfRangeException(
"v2", v2, "Vector v2 must be 3 dimensional!");
}
VectorC result = new VectorC(3);
result[0] = v1[1] * v2[2] - v1[2] * v2[1];
result[1] = v1[2] * v2[0] - v1[0] * v2[2];
result[2] = v1[0] * v2[1] - v1[1] * v2[0];
return result;
}
public MatrixC ReplaceRow(VectorC v, int n)
{
if (n < 0 || n > Rows)
{
throw new ArgumentOutOfRangeException(
"n", n, "n is out of range!");
}
if (v.GetSize() != Cols)
{
throw new ArgumentOutOfRangeException(
"Vector size", v.GetSize(), "vector size is out of range!");
}
for (int i = 0; i < Cols; i++)
{
matrix[n, i] = v[i];
}
return new MatrixC(matrix);
}
public VectorC GetRowVector(int n)
{
if (n < 0 || n > Rows)
{
throw new ArgumentOutOfRangeException(
"n", n, "n is out of range!");
}
VectorC v = new VectorC(Cols);
for (int i = 0; i < Cols; i++)
{
v[i] = matrix[n, i];
}
return v;
}
public static MatrixC Transform(VectorC v1, VectorC v2)
{
if (v1.GetSize() != v2.GetSize())
{
throw new ArgumentOutOfRangeException(
"v1", v1.GetSize(), "The vectors must have the same size!");
}
MatrixC result = new MatrixC(v1.GetSize(), v1.GetSize());
for (int i = 0; i < v1.GetSize(); i++)
{
for (int j = 0; j < v1.GetSize(); j++)
{
result[j, i] = v1[i] * v2[j];
}
}
return result;
}
public static VectorC Transform(VectorC v, MatrixC m)
{
VectorC result = new VectorC(v.GetSize());
//if (!m.IsSquared())
//{
// throw new ArgumentOutOfRangeException(
// "Dimension", m.GetRows(), "The matrix must be squared!");
//}
//if (m.GetRows() != v.GetSize())
//{
// throw new ArgumentOutOfRangeException(
// "Size", v.GetSize(), "The size of the vector must be equal"
// + "to the number of rows of the matrix!");
//}
for (int i = 0; i < m.GetRows(); i++)
{
result[i] = 0.0;
for (int j = 0; j < m.GetCols(); j++)
{
result[i] += v[j] * m[j, i];
}
}
return result;
}
