v0.11.0
github-actions
released this
08 Apr 15:54
·
133 commits
to master
since this release
GeometricFlux v0.11.0
- Adds GATv2 layer
- add DeepSet model and digit sum example
- fix GAT example
- add tutorials
- replace
Flux.destructure
byOptimisers.destructure
Closed issues:
- Lack of documentations and tutorials (#3)
- GAT example fails (#43)
- Edge features in GCNConv layer (#77)
- GAT example not working (#135)
- GAT outputs same embedding for all nodes if defined on complete graph (#154)
- Bug with CUDA + GATConv (#185)
- Add examples to FluxBench.jl (#228)
- Implementation of DeepSets (#237)
- Implementation of Set2set (#238)
- gradient for GATConv layer is very slow (#243)
- How do you use
bypass_graph
with aFlux.Chain
? (#248) - examples/gcn.jl not working (#262)
Merged pull requests:
- Add tests for Gatv2 layer (#267) (@yuehhua)
- Add DeepSet model and example (#278) (@yuehhua)
- Replace Flux.destructure by Optimisers.destructure (#279) (@yuehhua)
- Fix GAT example (#280) (@yuehhua)
- Memory pre-allocation fix for GAT example (#282) (@yuehhua)
- Update docs and fix GATv2Conv (#283) (@yuehhua)