public static TensorOld <double> Dot(this TensorOld <double> t1, TensorOld <double> t2)
{
if (t1.NumberDimensions > 2 || t2.NumberDimensions > 2)
{
throw new ProgrammingError("Dot product is not defined for tensors more that dimension 2");
}
if (t1.NumberDimensions == 1)
{
t1 = t1.Reshape(t1.Len, 1);
}
if (t2.NumberDimensions == 1)
{
t2.Reshape(1, t2.Len);
}
if (t1.ShapeDimensions[1] != t2.ShapeDimensions[0])
{
throw new ProgrammingError($"number of colums of the first matrix '{t1.ShapeDimensions[1]}' is not equal to the number of rows of the second: '{t2.ShapeDimensions[0]}'");
}
int convolutionSize = t1.ShapeDimensions[1];
TensorOld <double> result = new TensorOld <double>(new[] { t1.ShapeDimensions[0], t2.ShapeDimensions[1] });
int numRows = t1.ShapeDimensions[0];
int numCols = t2.ShapeDimensions[1];
for (int col = 0; col < numCols; col++)
{
result.GetTensorSpan(1, col);
for (int row = 0; row < numRows; row++)
{
ArraySubCollection <double> iter1 = t1.GetTensorSpan(1, row);
ArraySubCollection <double> iter2 = t2.GetTensorSpan(0, 0, col);
double cellValue = 0;
for (int i = 0; i < convolutionSize; i++)
{
cellValue += iter1[i] * iter2[i];
}
}
}
return(result);
}
private TensorOld GetUnFlatten(TensorOld error)
{
return(error.Reshape(lastInputShape));
}
private TensorOld GetFlatten(TensorOld input)
{
var d1 = input.shape[0];
return(input.Reshape(d1, input.ElementCount / d1));
}
