GitLab

* Uses the `trust_data` invariant to skip blocking assertions, when unnecessary, during construction of `SparseStorage` objects.
* Refactors the dtype and device transfer APIs to align with `torch.Tensor` while maintaining backward compatibility.
* No longer constructs dummy tensors when changing dtype or device.

* fix(`SparseTensor.__getitem__`): support `np.ndarray` and fix `List[bool]`

support indexing with np.ndarray & fix bug merging from indexing with
List[bool]

* style(tensor, test_tensor): pep8 E501 too long

support indexing with np.ndarray & fix bug merging from indexing
with
List[bool]

* update

* typo

* typo
Co-authored-by: tim <dbwtimteo@outlook.com>
Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

* Create spadd.py

Hi,
Maybe it's trivial to have this function, but I still think it'll be helpful and it looks neat when applying matrix addition, i.e., C = A + B.
Thanks

* update

* update

* fix jit
Co-authored-by: rusty1s <matthias.fey@tu-dortmund.de>

BUG: when sparse_sizes is passed with (None, None), this causes an error

a simple modification to the code of spmm() can make it support also 3D tensor (might think as batched):
suppose dense matrix has dimensions (B, N, F), the nonzero has dimensions (B, num_edges), all graphs share the same topology (i.e., the same adjacency matrix)