__init__.py 文件源码

def to_sparse(dense):
    mask = dense.ne(0)
    indices = mask.nonzero()
    if indices.storage():
        values = dense[mask]
    else:
        indices = indices.resize_(1, dense.ndimension()).zero_()
        values = dense.new().resize_(1).zero_()

    # Construct sparse tensor
    klass = getattr(torch.sparse, dense.__class__.__name__)
    res = klass(indices.t(), values, dense.size())
    if dense.is_cuda:
        res = res.cuda()
    return res