rusty1s/pytorch_cluster

random_walk result not reproducible if the walk is biased

jsun57 opened this issue · 8 comments

As title. The issue comes from

time(NULL), rowptr.data_ptr<int64_t>(), col.data_ptr<int64_t>(),
,
where the seed is controlled by time(NULL) in rejection_sampling_kernel function. Please fix~

Can you send a PR to fix this?

While I can locate the issue, unfortunately I only have access to a university cluster which does not have nvcc installed... So I am not able to build from source to perform testing sadly

Ok, I see if I can find time to implement this.

Wondering if there are any updates on this