internal T DeterminantLaplace(int diagLength)
{
if (diagLength == 1)
{
return(ConstantsAndFunctions <T> .Forward(this.GetValueNoCheck(0, 0)));
}
var det = ConstantsAndFunctions <T> .CreateZero();
var sign = ConstantsAndFunctions <T> .CreateOne();
var temp = SquareMatrixFactory <T> .GetMatrix(diagLength);
for (int i = 0; i < diagLength; i++)
{
GetCofactor(this, temp, 0, i, diagLength);
det = ConstantsAndFunctions <T> .Add(det,
ConstantsAndFunctions <T> .Multiply(
sign,
ConstantsAndFunctions <T> .Multiply(
this.GetValueNoCheck(0, i),
temp.DeterminantLaplace(diagLength - 1)
))
);
sign = ConstantsAndFunctions <T> .Negate(sign);
}
return(det);
}
private GenTensor <SafeDivisionWrapper <T> > InnerGaussianEliminationSafeDivision(int n)
{
InitIfNotInitted();
var elemMatrix = GenTensor <SafeDivisionWrapper <T> >
.CreateMatrix(n, n,
(x, y) => new SafeDivisionWrapper <T>(ConstantsAndFunctions <T> .Forward(this.GetValueNoCheck(x, y)))
);
for (int k = 1; k < n; k++)
{
for (int j = k; j < n; j++)
{
var m = ConstantsAndFunctions <SafeDivisionWrapper <T> > .Divide(
elemMatrix.GetValueNoCheck(j, k - 1),
elemMatrix.GetValueNoCheck(k - 1, k - 1)
);
for (int i = 0; i < n; i++)
{
var curr = elemMatrix.GetValueNoCheck(j, i);
elemMatrix.SetValueNoCheck(ConstantsAndFunctions <SafeDivisionWrapper <T> > .Subtract(
curr,
ConstantsAndFunctions <SafeDivisionWrapper <T> > .Multiply(
m,
elemMatrix.GetValueNoCheck(k - 1, i)
)
), j, i);
}
}
}
return(elemMatrix);
}
internal void Assign(GenTensor <T> genTensor)
{
foreach (var(index, value) in genTensor.Iterate())
{
this.SetValueNoCheck(ConstantsAndFunctions <T> .Forward(value), index);
}
}
/// <summary>
/// You might need it to make sure you don't copy
/// your data but recreate a wrapper (if have one)
///
/// O(V)
/// </summary>
public GenTensor <T> Forward()
{
var res = new GenTensor <T>(Shape);
foreach (var index in res.IterateOverElements())
{
res.SetValueNoCheck(ConstantsAndFunctions <T> .Forward(GetValueNoCheck(index)), index);
}
return(res);
}
/// <summary>
/// [i, j, k...]th element of the resulting tensor is
/// operation(a[i, j, k...], b[i, j, k...])
/// </summary>
public static GenTensor <T> Zip(GenTensor <T> a,
GenTensor <T> b, Func <T, T, T> operation)
{
#if ALLOW_EXCEPTIONS
if (a.Shape != b.Shape)
{
throw new InvalidShapeException("Arguments should be of the same shape");
}
#endif
var res = new GenTensor <T>(a.Shape);
if (res.Shape.shape.Length == 1)
{
for (int x = 0; x < res.Shape.shape[0]; x++)
{
res.Data[x] = ConstantsAndFunctions <T> .Forward(
operation(a.GetValueNoCheck(x), b.GetValueNoCheck(x)));
}
}
else if (res.Shape.shape.Length == 2)
{
for (int x = 0; x < res.Shape.shape[0]; x++)
{
for (int y = 0; y < res.Shape.shape[1]; y++)
{
res.Data[x * res.Blocks[0] + y] = ConstantsAndFunctions <T> .Forward(
operation(a.GetValueNoCheck(x, y), b.GetValueNoCheck(x, y)));
}
}
}
else if (res.Shape.shape.Length == 3)
{
for (int x = 0; x < res.Shape.shape[0]; x++)
{
for (int y = 0; y < res.Shape.shape[1]; y++)
{
for (int z = 0; z < res.Shape.shape[2]; z++)
{
res.Data[x * res.Blocks[0] + y * res.Blocks[1] + z] = ConstantsAndFunctions <T> .Forward(
operation(a.GetValueNoCheck(x, y, z), b.GetValueNoCheck(x, y, z)));
}
}
}
}
else
{
foreach (var index in res.IterateOverElements())
{
res.SetValueNoCheck(ConstantsAndFunctions <T> .Forward(
operation(a.GetValueNoCheck(index), b.GetValueNoCheck(index))), index);
}
}
return(res);
}
// TODO: how to avoid code duplication?
/// <summary>
/// Performs simple Gaussian elimination method on a tensor
///
/// O(N^3)
/// </summary>
public T DeterminantGaussianSimple()
{
#if ALLOW_EXCEPTIONS
if (!IsMatrix)
{
throw new InvalidShapeException("this should be matrix");
}
if (Shape[0] != Shape[1])
{
throw new InvalidShapeException("this should be square matrix");
}
#endif
if (Shape[0] == 1)
{
return(ConstantsAndFunctions <T> .Forward(this.GetValueNoCheck(0, 0)));
}
var n = Shape[0];
var elemMatrix = this.Forward();
for (int k = 1; k < n; k++)
{
for (int j = k; j < n; j++)
{
var m = ConstantsAndFunctions <T> .Divide(
ConstantsAndFunctions <T> .Forward(elemMatrix.GetValueNoCheck(j, k - 1)),
ConstantsAndFunctions <T> .Forward(elemMatrix.GetValueNoCheck(k - 1, k - 1))
);
for (int i = 0; i < n; i++)
{
var curr = ConstantsAndFunctions <T> .Forward(elemMatrix.GetValueNoCheck(j, i));
elemMatrix.SetValueNoCheck(ConstantsAndFunctions <T> .Subtract(
curr,
ConstantsAndFunctions <T> .Multiply(
m,
elemMatrix.GetValueNoCheck(k - 1, i)
)
), j, i);
}
}
}
var det = ConstantsAndFunctions <T> .CreateOne();
for (int i = 0; i < n; i++)
{
det = ConstantsAndFunctions <T> .Multiply(det, elemMatrix.GetValueNoCheck(i, i));
}
return(det);
}
public static GenTensor <T> Concat(GenTensor <T> a, GenTensor <T> b)
{
#if ALLOW_EXCEPTIONS
if (a.Shape.SubShape(1, 0) != b.Shape.SubShape(1, 0))
{
throw new InvalidShapeException("Excluding the first dimension, all others should match");
}
#endif
if (a.IsVector)
{
var resultingVector = GenTensor <T> .CreateVector(a.Shape.shape[0] + b.Shape.shape[0]);
for (int i = 0; i < a.Shape.shape[0]; i++)
{
resultingVector.SetValueNoCheck(ConstantsAndFunctions <T> .Forward(a.GetValueNoCheck(i)), i);
}
for (int i = 0; i < b.Shape.shape[0]; i++)
{
resultingVector.SetValueNoCheck(ConstantsAndFunctions <T> .Forward(b.GetValueNoCheck(i)), i + a.Shape.shape[0]);
}
return(resultingVector);
}
else
{
var newShape = a.Shape.Copy();
newShape.shape[0] = a.Shape.shape[0] + b.Shape.shape[0];
var res = new GenTensor <T>(newShape);
for (int i = 0; i < a.Shape.shape[0]; i++)
{
res.SetSubtensor(a.GetSubtensor(i), i);
}
for (int i = 0; i < b.Shape.shape[0]; i++)
{
res.SetSubtensor(b.GetSubtensor(i), i + a.Shape.shape[0]);
}
return(res);
}
}
/// <summary>
/// Copies a tensor calling each cell with a .Copy()
///
/// O(V)
/// </summary>
public GenTensor <T> Copy(bool copyElements)
{
var res = new GenTensor <T>(Shape);
if (!copyElements)
{
foreach (var index in res.IterateOverElements())
{
res.SetValueNoCheck(ConstantsAndFunctions <T> .Forward(GetValueNoCheck(index)), index);
}
}
else
{
foreach (var index in res.IterateOverElements())
{
res.SetValueNoCheck(ConstantsAndFunctions <T> .Copy(GetValueNoCheck(index)), index);
}
}
return(res);
}
/// <summary>
/// Finds Determinant with possible overflow
/// because it uses fractions for avoiding division
///
/// O(N^3)
/// </summary>
internal T DeterminantGaussianSafeDivision(int diagLength)
{
InitIfNotInitted();
#if ALLOW_EXCEPTIONS
if (!IsMatrix)
{
throw new InvalidShapeException("this should be matrix");
}
if (Shape[0] != Shape[1])
{
throw new InvalidShapeException("this should be square matrix");
}
#endif
if (Shape[0] == 1)
{
return(ConstantsAndFunctions <T> .Forward(this.GetValueNoCheck(0, 0)));
}
var n = diagLength;
var elemMatrix = InnerGaussianEliminationSafeDivision(n);
var det =
ConstantsAndFunctions <SafeDivisionWrapper <T> > .CreateOne();
for (int i = 0; i < n; i++)
{
det = ConstantsAndFunctions <SafeDivisionWrapper <T> > .Multiply(det, elemMatrix.GetValueNoCheck(i, i));
}
if (ConstantsAndFunctions <T> .IsZero(det.den))
{
return(ConstantsAndFunctions <T> .CreateZero());
}
return(det.Count());
}
