I would like to use sparse matrices in diffusion models.
sp is my Minkowski tensor, and I need to frequently add noise to sp.
For Minkowski Engine, it seems that it does not support direct modification of data features. Therefore, I need to frequently call the constructor function:
sp = ME.SparseTensor(features=sp.F+noise, coordinate_manager=sp.coordinate_manager, coordinate_map_key=sp.coordinate_map_key)
Each call takes a significant amount of time to allocate memory. Are there any good solutions to this issue?
I would like to use sparse matrices in diffusion models. sp is my Minkowski tensor, and I need to frequently add noise to sp. For Minkowski Engine, it seems that it does not support direct modification of data features. Therefore, I need to frequently call the constructor function:
sp = ME.SparseTensor(features=sp.F+noise, coordinate_manager=sp.coordinate_manager, coordinate_map_key=sp.coordinate_map_key)Each call takes a significant amount of time to allocate memory. Are there any good solutions to this issue?