to_scipy

iskra.sparse.to_scipy(x: SparseTensor | Tensor) → coo_array | csr_array[SOURCE]

Converts PyTorch sprase COO/CSR tensors into SciPy sparse arrays.

If the input is on the CPU and does not require gradients, we will return a view into the data. If the data is on the GPU or requires gradients, the tensor will be cloned and a view into that tensor will be returned.

Since SciPy does not perform type erasure when it comes to sparse layouts, this function can only return the type scipy.sparse.coo_array | scipy.sparse.csr_array, which might make your type checker complain if you call any functions specific to either layout. If this happens, you can always follow this function up with, e.g., assert isinstance(y, scipy.sparse.csr_array) to silence the type checker.

Parameters:

x (SparseTensor | Tensor) – COO or CSR tensor to be converted.

Raises:

ValueError – Throws an exception if the input tensor is neither COO nor CSR.

Returns:

​ (scipy.sparse.coo_array | scipy.sparse.csr_array) – SciPy array with the same data.