Get initial benchmarks across all pooling on a few datasets

bayesmech / research-gnn-pooling

Research on Heirarchical pooling for Graph Neural Networks

MIT License

0 stars 0 forks source link

Get initial benchmarks across all pooling on a few datasets #11

Closed AnimeshSinha1309 closed 2 years ago

AnimeshSinha1309 commented 2 years ago

We need to run tests on how our model performs against all the other pooling methods there. For this run, we choose to benchmark against:

DiffPool - Because it's most similar to us and we should be somewhere close to it, hopefully, more memory efficient and better as we go deeper.
ASAPool - Because this class of node deletion methods is the current best competitor, with HGP-SL and others being on the same lines.

AnimeshSinha1309 commented 2 years ago

These are the results from the run: WandB Sweep

These paint a very optimistic picture as our model performs better than others on all fronts. We now need to validate if this is true as we vary backbone scale and other hyperparameters.