Pytorch Sparse Versions Save

Fixed a minor bug that caused torch-sparse to crash in PyTorch 1.7.0.

• PyTorch 1.6.0 wheels
• Fixed a bug for reductions in dim=0
• Fixed a bug in which PyTorch warnings were not displayed when importing torch-sparse

• cat can now concatenate a list of SparseTensors diagonally by passing dim=(0,1) to the function call.

• Better JIT support for matmul via @overload
• Replaced options arguments with device and dtype arguments

Added neighborhood sampling functionality via sample and sample_adj.

• Fixed a bug in spspmm on the CPU
• Added sparse_reshape functionality
• Added bandwidth utilities

This release introduces random walk and GraphSAINT subgraph functionalities via random_walk and saint_subgraph to the SparseTensor class.

This release introduces the partition function based on the METIS library which re-orders the entries of a SparseTensor according to a computed partition. Note that the METIS library needs to be installed and WITH_METIS=1 needs to be set in order to make use of this function.

This release includes a major rewriting of torch-sparse by introducing the SparseTensor class, which is fully differentiable and traceable. The SparseTensor class is still undocumented and not well-tested, and should hence be used with caution. All known functions from earlier versions still work as expected.

Support for torch-scatter=2.0. As a result, PyTorch 1.4 is now required to install this package.