cugraph
cugraph copied to clipboard
rapidsai
cuGraph - RAPIDS Graph Analytics Library
Implement Leiden algorithm. Implementation should use the graph primitives. Because of Leiden's similarity to Louvain, it is expected that the existing primitives should be sufficient to implement Leiden. If there...
reach out to JoC team and obtain their heterogeneous data generator. https://github.com/Nanthini10/synthetic-data Then create test datasets - 1M edges - 10M edges - 100M edges - 1,000M edges
**Describe the solution you'd like and any additional context** Currently in PG [add_edge_data](https://github.com/rapidsai/cugraph/blob/cc057589799366b44eacf8b6f4a335223e4927c2/python/cugraph/cugraph/structure/property_graph.py#L478-L505) we assign `edge_ids` on our own and the user cant add its own if they want to....
Does not include GaaS. Supports heterogeneous graphs and MG. Includes example notebook. Closes https://github.com/rapidsai/graph_dl/issues/20 Closes https://github.com/rapidsai/graph_dl/issues/21
Partially address #2479 We currently support only edge weight as edge properties. This PR's goal is to update cuGraph to support additional edge properties (currently only edge ID, eventually any...
**Describe the solution you'd like and any additional context** We need a primitive to gather information from each vertex's incoming/outgoing edges. This will be served with `per_v_transform_reduce_incoming|outgoing_e` if the gathered...
Define the C API for Leiden. Should be very similar to the C API for Louvain.
Implement MNMG Leiden. We have an SG implementation, although it does not accurately compute the Leiden clustering algorithm. The new implementation should be primitive based and more accurately reflect the...
Define C API unit tests. These do not need to duplicate the C++ tests but should focus on testing all different mechanisms on the C API. Tests should include both...
Define C++ unit tests for Leiden. Mimicking the basic tests for our algorithms might be sufficient. The tests should contain some tests where the clustering differs from Louvain, especially for...