Seunghwa Kang

> Otherwise we need an additional 72 explicit instantiations and I can address it in a separate PR. Why 72? I assume 3 (vertex_t & edge_t combinations) * 4 (store_transposed...

> We currently support heterogeneous renumbering in the C++ sampling API. Do we? What does heterogenous sampling mean for non-heterogeneous graph? > A heterogeneous graph can have multiple node types...

> Each edge type corresponds to a separate subgraph in the output minibatch. And I am not sure how our `compress_per_hop` option should work with this. If there are multiple...

@alexbarghi-nv https://github.com/rapidsai/cugraph/pull/4463 needs your review.

Yeah... and I think we need to think little more about whether this is possible or not. Say, to compute degrees, the maximum possible degree can be larger than the...

Yes, edge ID is one thing (for this, we can introduce edge_id_t) and returning degrees is another. This being said, we still don't need to use edge_t for the offsets...

I guess for both (and eventually, we may better use the same data set for Python and C++ testing). The main rationale behind this is to test for multiple corner...

And if we can cover sufficient corner cases with tiny graphs, we may skip testing with large files (which takes long to read & parse and increase our C++ testing...