[BUG] Fix Issue in cuGraph-PyG Tests Blocking CI #3607
Conversation
alexbarghi-nv
added
bug
…-nv/cugraph into fix-loader-ci-failure
…rkaround a numba 0.57 incompatibility.
…nd a breaking cudf change related to its numba dependency.
| @@ -79,7 +78,7 @@ def karate_gnn(): | |||
| el = karate.get_edgelist().reset_index(drop=True) | |||
| el.src = el.src.astype("int64") | |||
| el.dst = el.dst.astype("int64") | |||
| all_vertices = np.array_split(cudf.concat([el.src, el.dst]).unique().values_host, 2) | |||
| all_vertices = np.array_split(np.arange(34), 2) | |||
There was a problem hiding this comment.
This failure is probably caused by rapidsai/cudf#11656.
The ordering of cudf's unique and drop_duplicates has changed. Previously, the order was sorted. Now, the order is stable (preserves input order while removing duplicates). This may affect other parts of cuGraph, too, so I will take a look.
There was a problem hiding this comment.
Right, I was aware of that. Ironically, I have another PR which takes advantage of the new default behavior for drop_duplicates.
There was a problem hiding this comment.
Yes, it looks like unique is heavily used in cuGraph's Python code. We may need to review these for correctness if sorted order is expected. https://github.com/search?q=repo%3Arapidsai%2Fcugraph+unique%28+language%3APython&type=code&l=Python
There was a problem hiding this comment.
Awesome, thanks @alexbarghi-nv! Glad to hear the new behavior is helpful for your other PR! :D
There was a problem hiding this comment.
Approving doc updates from the https://github.com/bdice/cugraph/tree/numba-0.57 branch that had to be merged in.
|
/merge |
One of the fixtures for the cuGraph-PyG tests is not properly partitioning the edgelist, causing some tests to fail. This PR updates that fixture to correctly partition the edgelist.