v0.7.1

GeometricFlux v0.7.1

Diff since v0.7.0

• Add GraphMLDatasets as dependency to provide datasets

Closed issues:

• Provide ready-to-use datasets (#4)

Merged pull requests:

• Add datasets (#98) (@yuehhua)
• Add GraphMLDatasets as deps (#101) (@yuehhua)

v0.7.0

GeometricFlux v0.7.0

Diff since v0.6.3

Merged pull requests:

• Correct VGAE doc (#88) (@yuehhua)
• Correct VGAE example (#89) (@yuehhua)
• Allow DataStructures 0.18 (#90) (@ChrisRackauckas)
• Make CompatHelper use latest Julia (#91) (@ChrisRackauckas)
• Integrated datasets (#92) (@yuehhua)

v0.6.3

GeometricFlux v0.6.3

Diff since v0.6.2

• GDE, GAE VGAE examples available
• Correct GCNConv show

Closed issues:

• VGAE implementation (#9)

Merged pull requests:

• Neural Graph Differential Equations example works on CPU (#81) (@yuehhua)
• Correct GCNConv show (#82) (@yuehhua)
• GDE example works on GPU (#83) (@yuehhua)
• Generic algorithm for GCNConv (#84) (@yuehhua)
• VGAE example (#85) (@yuehhua)
• Correct GAE example (#86) (@yuehhua)
• Add doc for GAE and VGAE (#87) (@yuehhua)

v0.6.2

GeometricFlux v0.6.2

Diff since v0.6.1

• Add FeatureSelector
• Correct ChebConv computation
• Make scaled_laplacian differentiable
• Add ScatterNNlib and GraphSignals as deps
• Improve GAT example
• Upgrade to CUDA
• Maintain Travis CI

Closed issues:

• Metagraphs.jl integraion (#6)
• LightGraphs dependency warning (#24)
• Support for graph nets (#36)
• Sound support of variable graphs (#42)

Merged pull requests:

• Allow ci test nightly (#70) (@yuehhua)
• Fix differentiability of layers (#71) (@yuehhua)
• Add FeatureSelector (#78) (@yuehhua)
• Remove Reference from FeaturedGraph field (#79) (@yuehhua)
• Add ScatterNNlib and GraphSignals as deps (#80) (@yuehhua)

v0.6.1

GeometricFlux v0.6.1

Diff since v0.6.0

• Update to CUDA 1.2 and Flux 0.11
• Refactor graph-related API
• Improve learning rate in example

Merged pull requests:

• GAE example (#65) (@yuehhua)
• CompatHelper: bump compat for "Flux" to "0.11" (#67) (@github-actions[bot])
• Update to CUDA.jl 1.2 and Flux 0.11 (#68) (@yuehhua)
• Graph API enhancements (#69) (@yuehhua)

v0.6.0

GeometricFlux v0.6.0

Diff since v0.5.2

• Rewrite graph network GraphNet and message passing MessagePassing framework
• Expand functionality of FeaturedGraph to support node_feature, edge_feature and global_feature
• Speed up ChebConv layer
• Speed up scatter functions
• Add graph index-related functions
• GCN example works and increase training stablility
• Fix show GCNConv
• Add more test for linear algebra
• Update cpu scatter benchmark plot and scripts

Closed issues:

• References in FeaturedGraphs (#58)
• Define @adjoint on _apply_msg! (#59)
• GCNConv example fails (#61)

Merged pull requests:

• Try to fix GAT example by split utility dependency for SimpleGraph and SimpleWeightedGraphs (#45) (@ilancoulon)
• Improve CPU scatter performance (#56) (@yuehhua)
• Ignore gradient of generate_cluster (#57) (@yuehhua)
• Fix reference bug (#60) (@yuehhua)
• Rewrite GraphNet and MessagePassing framework (#63) (@yuehhua)
• GCN example works again (#64) (@yuehhua)

v0.5.2

GeometricFlux v0.5.2

Diff since v0.5.1

• Add scaled Laplacian
• Support CuArrays v2.0 and Flux v0.10.4
• ChebConv, GraphConv, GATConv, GatedGraphConv and EdgeConv support FeaturedGraph
• Add SimpleWeightedGraphs and MetaGraphs as deps
• Fix broadcastly casting error

Merged pull requests:

• Variable graphs (#44) (@ilancoulon)
• CompatHelper: bump compat for "IRTools" to "0.4" (#46) (@github-actions[bot])
• Fix bug while update to CuArrays v2.0.0 (#48) (@yuehhua)
• CompatHelper: bump compat for "Zygote" to "0.5" (#49) (@github-actions[bot])
• Support latest CUDA-related packages (#50) (@yuehhua)
• Support scaled Laplacian for other graphs (#51) (@yuehhua)
• CompatHelper: add new compat entry for "MetaGraphs" at version "0.6" (#52) (@github-actions[bot])
• CompatHelper: add new compat entry for "SimpleWeightedGraphs" at version "1.1" (#53) (@github-actions[bot])

v0.5.1

GeometricFlux v0.5.1

Diff since v0.5.0

• GCNConv layer supports FeaturedGraph (#34)
• Support linear algebra for FeaturedGraph
• Add nv API for FeaturedGraph
• Add LightGraphs as dependency
• Correct normalized laplacian type
• Fix bug in normalized_laplacian
• Fix Base.show on GCNConv
• Add docs (#35)

Closed issues:

• Test errors on Julia 1.3 and 1.4 master (#14)

Merged pull requests:

• GCNConv layer supports FeaturedGraph (#34) (@yuehhua)
• Add docs (#35) (@yuehhua)
• Fix Base.show on GCNConv and GCNConv output FeaturedGraph (#38) (@yuehhua)
• Fix bug (#39) (@yuehhua)
• Update docs (#40) (@yuehhua)
• Some updates to get the GCN example working... (#41) (@kshyatt)
• Support linear algebra for FeaturedGraph and some refactoring (#47) (@yuehhua)

v0.5.0

GeometricFlux v0.5.0

Diff since v0.4.0

• Support scatter operations for MArray (#32)
• Support GCNConv layer accepting graph input (#31)

Merged pull requests:

• Don't need atomic operations (#30) (@yuehhua)
• Support layer accepting graph input (#31) (@yuehhua)
• Support scatter operations for MArray (#32) (@yuehhua)

v0.4.0

GeometricFlux v0.4.0

Diff since v0.3.0

• Compatible with Julia v1.4 while not support before v1.3
• Not support old version CuArrays, CUDAnative and CUDAapi
• Improve performance of scatter operations for CPU and new benchmark (#29)
• Scatters support almost all Real numbers except Bool on CPU
• Add benchmark for scatter operations
• Implement TopKPool layer (#22)

Closed issues:

• Top-k pooling layer (#5)

Merged pull requests:

• Scatters support almost all Real numbers and add test coverage (#21) (@yuehhua)
• Implement TopKPool layer (#22) (@yuehhua)
• CompatHelper: bump compat for "CUDAapi" to "4.0" (#23) (@github-actions[bot])
• CompatHelper: bump compat for "CUDAnative" to "3.0" (#25) (@github-actions[bot])
• CompatHelper: bump compat for "CuArrays" to "2.0" (#26) (@github-actions[bot])
• Update Julia to 1.4 in Travis (#27) (@XVilka)
• Not support old version of CuArrays, CUDAnative and CUDAapi (#28) (@yuehhua)
• Use view in scatter for CPU and new benchmark (#29) (@yuehhua)