Unify the graph representation

CemOezcan / hyper-graph-nets

Implementation of different remote message passing strategies in graph neural networks for mesh-based physical simulation.

MIT License

6 stars 0 forks source link

Open CemOezcan opened 1 year ago

CemOezcan commented 1 year ago

[ ] Remove computation of unnecessary graph components (such as node_dynamics)
[ ] Decouple feature normalization from graph computation
[ ] Keep unused instances of a trajectory on the cpu