[QST]: Should nx-cugraph GPU graph caching be opt-out rather than opt-in?

[beckernick]

What is your question?

Converting a CPU NetworkX graph to a GPU graph can be slow, so nx-cuGraph provides a configuration option that caches the converted graph so users only need to pay this cost once regardless of how many algorithms they run on the graph (NETWORKX_CACHE_CONVERTED_GRAPHS=True).

This feels high-impact if users frequently run multiple algorithms per graph. Is there any reason we wouldn't want to turn this on by default when someone is using nx-cuGraph? E.g., are the graphs people frequently use with NetworkX algorithms particularly large in terms of memory? Or, do we expect people to be manually mutating the graphs?

Code of Conduct

• [X] I agree to follow cuGraph's Code of Conduct
• [X] I have searched the open issues and have found no duplicates for this question