Sparse matrix modification

I am working with sparse matrix and I very frequently modify my matrix by adding new elements or changing its size. How to do it efficiently ?
Because, at the moment, each time I do a modification I use again the constructor torch.sparse.FloatTensor(indices, values, size) and I am also obliged to use the method coalesce (because I am using indices() and values() ) even if I know that my matrix is already coalesced.

How can I simplify and accelerate the modification of my sparse matrix ?
Thank you