cugraph
cugraph copied to clipboard
Create a mechanism to take multi-graph and generate a simple graph
Based on conversations with GNN developers, we need a mechanism that will allow us to take a multi-graph and generate a simple graph from it, with a user-provided function to transform the multi-edges into a single edge.
For example, you might sum the weights when collapsing an edge. Or you might take the maximum weight when collapsing an edge.
There are GNN use cases where sampling on a multi-graph will be done repeatedly and they don't want to get duplicate edges from the multi-graph. This would allow for the computation to be more efficient.
Need to consider whether this should be done in python or in C++.
My intuition is C++, as we already have to shuffle the edges around to create a graph, we might as well do that computation on the C++ side. We will probably start with simple cases (sum and max) rather than supporting an arbitrary UDF (user defined function).
This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.
Duplicated by #4057