Open sunitj opened 11 months ago
Thanks for your interest!
We will review the linked papers and consider how this could be done in cuGraph and update this issue with information on where it might fit in our road map.
I am also very much interested in an easily accessible GPU-accelerated MCL implementation!
Is this a new feature, an improvement, or a change to existing functionality?
New Feature
How would you describe the priority of this feature request
Critical (currently preventing usage)
Please provide a clear description of problem this feature solves
I would like cuGraph to implement the Markov clustering algorithm, because I would like to perform such a clustering on my graphs where, nodes are proteins, edges denote similarity between the nodes and edge weights are similarities. MCL is a commonly used algorithm in analyzing protein similarity networks. A recent publication performed such a clustering on a graph that consisted of
570,198,677
nodes and5,196,499,560
edges using a distributed version of the algorithm, HipMCL. They required2,500
compute nodes (170,000
compute cores) on the NERSC super computer for3h20m
. I estimate my graph to be around the same order of nodes and edges (if not more), but I don't have access to a supercomputer. I do have access to AWS and GPU instances. I should also add that I am completely new to the RAPIDS universe of tools, but quite excited and eager to learn.Describe your ideal solution
The new function takes an existing graph as described above. The function would perform the following steps (borrowed from this repo)
Describe any alternatives you have considered
I have considered:
but none of them have GPU acceleration and I don't have the resources to rent the NERSC supercomputer for 4 hours.
Additional context
Code of Conduct