TWrap

TWrap is a C++ wrapper class around the Eigen3-unsupported Tensor Module.

It is built via

mkdir build && cd build
cmake ..
make

A template executable - illustrating some core functionality - is provided and can be executed as

cd build
./bin/template

Unit tests can be run after compiling using ctest.

How to use TWrap?

• create a tensor

  WTens<double, 1> myTensor(5);
  

  creates a new Tensor with dimension 1 and length 5. You need to give as many length arguments as dimensions, e.g.

  WTens<int, 4> myTensor4D(1,2,3,4);
  

  All its elements are initialized to zero.

• set tensor elements

  Individual tensor values are set via

  myTensor(0) = 1.2;
  

  or you can set the whole tensor content using a C++ initializer_list

  myTensor.setValues({1,2,3,4,5});
  

  There are also the options to set all values to zero

  myTensor.setZero();
  

  or to a random value

  myTensor.setRandom();
  

• tensor operations

  • addition of tensors

  Two tensors are added via

  auto res = myTensor1 + myTensor2;
  

  • rescale a tensor with a factor

  By multiplying a tensor with a factor, all its content gets rescaled individually

  double scale = 2.0;
  auto res = myTensor * scale;
  

  • tensor multiplication

  Two tensors are multiplied using the concat function. You need to specify the indices which will be contracted. Below you see the example for a matix-vector multiplication.

  WTens<double, 1> vec(2);
  WTens<double, 2> matrix(2, 2);
  vec.setValues({10, 11});
  matrix.setValues({{1, 2}, {3, 4}});
  
  auto res = matrix.concat(vec, 1, 0);
  auto resT = matrix.concat(vec, 0, 0);
  

  res is the result for A * b, in indices: A_ij * b_j = c_i. resT the result for the product with the transposed vector, b^T * A, e.g. A_ij * b_i = c_j.