ProGraML icon indicating copy to clipboard operation
ProGraML copied to clipboard
verified publisher icon ChrisCummins

Add code for learning over ProGraML graphs

Open ChrisCummins opened this issue 4 years ago • 2 comments

🚀 Feature

I stripped the code for learning over ProGraML graphs from the tasks/ directory a while back as it was over complicated and slower than it needed to be. This issue is to track adding a new reference GGNN implementation to ProGraML.

  • [ ] Add an example data loader for PyTorch.
  • [ ] Re-implement the node-level classification code for the dataflow experiments.
  • [ ] Re-implement the graph-level classification code for the devmap experiment.
  • [ ] Re-implement the graph-level classification code for the classifyapp experiment.

ChrisCummins avatar Sep 06 '21 13:09 ChrisCummins

Hi, I would like to know approximately when the learning part will be available. Thanks very much~

Mem2019 avatar Oct 21 '21 04:10 Mem2019

Probably not soon, unfortunately :-( I am too busy on my other commitments to find the time to work on this.

You're welcome to take a look at an older revision of the learning code, or you could use the existing API to prepare your graphs in networkx / DGL then follow one of the many excellent tutorials on graph learning to get started.

Cheers, Chris

ChrisCummins avatar Oct 25 '21 10:10 ChrisCummins