Distinguish between dense and sparse vectors about clsparse HOT 4 CLOSED

What kind of data format you have in mind regarding storing the sparse vector? If we introduce sparse vector format, we need to introduce new versions of our algorithms.

from clsparse.

Yes, that's right. I note that cuSPARSE provides both dense and sparse vector formats: http://docs.nvidia.com/cuda/cusparse/#vector-formats

Most of the L2 routines operate with dense vector formats, hence we supported them first but there must be use cases for sparse vectors too. Until we have solid use cases for the sparse format, they should be lower priority than the other work we are already doing, but we need to keep an eye on our API's and make sure they inter-operate/compatible with both formats. For instance, C has no function overloading, so if we support AXPY and dot product functions, they have to have unique names for dense and sparse operations.

from clsparse.

I totally agree. From my experience (CFD) I didn't find / met any particular case where sparse vector format will overcome the dense implementation. Maybe somewhere in Quantum Physics or elsewhere such need exists.

from clsparse.

I now think that we should rename our clsparseVector class to be cldenseVector, as it really really represents dense data. We have yet to create a true clsparseVector.

Also, I think we should rename clsparseDenseMatrix as just cldenseMatrix for the same reason.

from clsparse.