-
Notifications
You must be signed in to change notification settings - Fork 304
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[FEA] Neighborhood sampling (MFG, single-GPU) #1978
Comments
I would like to work on this. My plan is to first submit a PR with just the API changes, so that we can already start reviewing changes at the API level while I can continue working on the implementation. |
This issue has been labeled |
This pull request adds neighborhood sampling, as needed by GNN frameworks (DGL, PyTorch-Geometric). Since I did not hear back on most of the other issues that need to be addressed before this, I am continuing with my plan of first opening a PR with just the API. Once we agree on the final API, and once a minimal version of cugraph-ops is integrated, we can add the implementation of this API. In particular, for now I am suggesting that the sampling type is exposed in the public API (it does not exist yet in cugraph-ops since that has not been integrated yet). This must be decided ahead of sampling for best performance (either by the end user or some automatic heuristic on the original graph), which is why it makes sense to have as a separate parameter for this API. EDIT: link to issue #1978 Authors: - Matt Joux (https://github.com/MatthiasKohl) Approvers: - AJ Schmidt (https://github.com/ajschmidt8) - Robert Maynard (https://github.com/robertmaynard) - Andrei Schaffer (https://github.com/aschaffer) - Chuck Hastings (https://github.com/ChuckHastings) URL: #1982
This issue has been labeled |
should be fixed by #1982 |
Is your feature request related to a problem? Please describe.
In order to use cugraph in frameworks such as DGL or Pytorch-Geometric, we should add an implementation of neighborhood sampling with the output in COO or CSR format in order to have the frameworks construct an MFG based on it.
Describe the solution you'd like
This should only happen at the C++ API level for now (libcugraph).
There should be a new algorithm providing COO or CSR given an input graph, a sample size and an array of seed vertices.
Exact API to be defined.
Implementation will be provided by (closed source) cugraph-ops which is in the process of getting integrated.
The text was updated successfully, but these errors were encountered: