Open vermorel opened 7 years ago
With a BrainScript script no more complex than:
lr = ReduceSum( SquareError(X, Z) )
with X being a sparse input, I end-up with the following exception:
X
Inside File: Matrix.cpp Line: 2300 Function: Microsoft::MSR::CNTK::Matrix<float>::operator -= -> Feature Not Implemented. [CALL STACK] > Microsoft::MSR::CNTK::Matrix<float>:: operator-= - Microsoft::MSR::CNTK::Matrix<float>:: AssignDifferenceOf - Microsoft::MSR::CNTK::Matrix<float>:: __autoclassinit2 (x13) - Microsoft::MSR::CNTK::MatrixBase:: MatrixBase EXCEPTION occurred: Inside File: Matrix.cpp Line: 2300 Function: Microsoft::MSR::CNTK::Matrix<float>::operator -= -> Feature Not Implemented.
It appears, indeed, that neither += nor -= are implemented in the case of the sparse matrix (same for CPU and GPU).
+=
-=
Might be a duplicate of #2143
To make sure it is a duplicate issue of #2143
With a BrainScript script no more complex than:
with
X
being a sparse input, I end-up with the following exception:It appears, indeed, that neither
+=
nor-=
are implemented in the case of the sparse matrix (same for CPU and GPU).