v0.6.21

GraphNeuralNetworks v0.6.21

Diff since v0.6.20

Merged pull requests:

• Add DCGRU temporal layer (#448) (@aurorarossi)
• move mldatasets2gnngraph (#458) (@CarloLucibello)
• create GNNLux.jl package (#460) (@CarloLucibello)
• [GNNLux] GCNConv, ChebConv, GNNChain (#462) (@CarloLucibello)
• [GNNLux] more layers (#463) (@CarloLucibello)
• use GNNlib in GNN.jl (#464) (@CarloLucibello)
• [GNNlib] fix cuda ext (#465) (@CarloLucibello)
• tests for GNNlib (#466) (@CarloLucibello)
• fix docs (#467) (@CarloLucibello)
• [GNNLux] fix tests (#468) (@CarloLucibello)
• [GNNLux] more layers (#469) (@CarloLucibello)
• [GNNLux] TGCN temporal layer (#470) (@aurorarossi)
• [GNNLux] more layers pt. 3 (#471) (@CarloLucibello)
• [GNNGraphs] implement remove_edges(g, p) (#474) (@rbSparky)
• [GNNLux] Added SGConv (#475) (@rbSparky)
• fix dense test (#479) (@CarloLucibello)
• [GNNLux] Adding MegNetConv Layer (#480) (@rbSparky)
• rng instead of seed for rand_graph (#482) (@CarloLucibello)
• @functor ->@layer (#484) (@CarloLucibello)
• [GNNLux] Add A3TGCN temporal layer (#485) (@aurorarossi)
• [GNNLux] Add GConvLSTM, GConvGRU and DCGRU temporal layers (#487) (@aurorarossi)
• fix NNConv docs (#488) (@CarloLucibello)
• Add EvolveGCNO temporal layer (#489) (@aurorarossi)
• [GNNLux] updates for Lux v1.0 (#490) (@CarloLucibello)
• [GNNLux] Adding NNConv Layer (#491) (@rbSparky)
• [GNNLux] add GMMConv, ResGatedGraphConv (#494) (@CarloLucibello)

Closed issues:

• Question about temporal graph neural networks (#112)
• add show methods for WithGraph and GNNChain (#178)
• Dropout inside GATConv layer (#258)
• Graph classification: multiple graphs associated with a common label (#282)
• convenience feature setter (#284)
• @functor default for GNNLayers (#288)
• documentation proposal (#357)
• support Lux (#372)
• turn this into a monorepo (#433)
• use Flux.@layer instead of Flux.@functor (#452)
• random graph generators should take an rng instead of a seed (#459)
• Cannot create GNNGraph with unconnected nodes (#472)
• Implementation of recommender system based on GNN (#473)
• GNNs.jl's CI is failing for GRAPH_T = :dense (#476)
• GCNConv layer fails when the GNNGraph comes from an adjacency matrix (#486)

v0.6.20

GraphNeuralNetworks v0.6.20

Diff since v0.6.19

Merged pull requests:

• Added perturb_edges function (#423) (@rbSparky)
• Add drop_nodes transform (#426) (@rbSparky)
• Added Personalized PageRank Diffusion [ppr_diffusion function] (#427) (@rbSparky)
• Added TAGConv layer (#430) (@rbSparky)
• create GNNlib.jl (#432) (@CarloLucibello)
• Create Register.yml (#434) (@CarloLucibello)
• Add GConvLSTM temporal layer (#437) (@aurorarossi)
• Add GConvGRU temporal layer (#438) (@aurorarossi)
• Add DConv layer (#441) (@aurorarossi)
• CompatHelper: bump compat for KrylovKit to 0.8, (keep existing compat) (#442) (@github-actions[bot])
• Coloring refinement algorithm (#444) (@CarloLucibello)
• separate GNNGraphs from GNNlib (#446) (@CarloLucibello)
• Add possibility to pass weights to GCNConv (#447) (@aurorarossi)
• temporarily reintegrate GNNGraphs tests (#449) (@CarloLucibello)
• CI for GNNGraphs.jl (#451) (@CarloLucibello)
• make GraphNeuralNetworks.jl depend on GNNGraphs.jl (#453) (@CarloLucibello)
• drop_nodes -> remove_nodes (#454) (@CarloLucibello)
• fix docs (#455) (@CarloLucibello)

Closed issues:

• Maybe state difference with GeometricFlux.jl. (#435)

v0.6.19

GraphNeuralNetworks v0.6.19

Diff since v0.6.18

Merged pull requests:

• feat: Add GCNConv support for HeteroGraphConv (#367) (@askorupka)
• feat: support weights when generating from SimpleWeightedGraph (#371) (@askorupka)
• Add TemporalSnapshotsGNNgraph classification tutorial (#408) (@aurorarossi)
• Creating dropout functionality in the GATConv and GATv2Conv Layers (#411) (@achiverram28)
• chore: update docs (layers compatible with GNNHeterograph) (#413) (@askorupka)
• Added remove_edges function (#414) (@rbSparky)
• fix: remove SGConv GNNHeteroGraph support (#416) (@askorupka)
• fix: GCNConv support for GNNHeteroGraph (#417) (@askorupka)
• Added remove_nodes function (#420) (@rbSparky)
• CompatHelper: bump compat for KrylovKit to 0.7, (keep existing compat) (#421) (@github-actions[bot])
• Added TransformerConv usage example (#422) (@rbSparky)
• Fixes and more tests in remove_nodes function (#424) (@rbSparky)
• chore: Update docs for HeteroGNN support (#425) (@askorupka)
• Bump julia-actions/setup-julia from 1 to 2 (#431) (@dependabot[bot])

Closed issues:

• Weights not included in GNNGraph made from SimpleWeightedDiGraph (#85)

v0.6.18

GraphNeuralNetworks v0.6.18

Diff since v0.6.17

Merged pull requests:

• Add GATv2Conv support to HeteroGraphConv (#407) (@rbSparky)
• Remove duplicate of GINConv for TemporalSnapshotsGNNGraph (#409) (@aurorarossi)
• Fix ref in the docs (#410) (@aurorarossi)

Closed issues:

• Duplicated method definitions of GINConv (#406)

v0.6.17

GraphNeuralNetworks v0.6.17

Diff since v0.6.16

Merged pull requests:

• feat: Add empty constructor for GNNHeteroGraph (#358) (@askorupka)
• Bump actions/cache from 3 to 4 (#359) (@dependabot[bot])
• feat: add degree functionality for GNNHeteroGraph (#360) (@askorupka)
• Fix ref rand_temporal_hyperbolic_graph (#361) (@aurorarossi)
• Add Flux.gpu function for TemporalSnapshotsGNNGraph type (#362) (@aurorarossi)
• feat: Add CGConv support for HeteroGraphConv (#363) (@askorupka)
• feat: Add EdgeConv support for HeteroGraphConv (#364) (@askorupka)
• Remove f from CGConv signature (#365) (@aurorarossi)
• fix: Stylistic fixes (#366) (@askorupka)
• fix: fix typos in the docs (#368) (@askorupka)
• chore: optimize test for heterograph (#370) (@askorupka)
• refinement: Self loops for HeteroGraph returns g instead of error if src != tgt (#373) (@askorupka)
• Bump codecov/codecov-action from 3 to 4 (#374) (@dependabot[bot])
• Adapt GINConv to TemporalSnapshotsGNNGraphs (#376) (@aurorarossi)
• Fix link ref Heterographs (#378) (@aurorarossi)
• Update documentation for convolutions on TemporalSnapshotsGNNGraphs (#379) (@aurorarossi)
• g.ndata.e documentation fix in GNNGraph (#381) (@rbSparky)
• Add SGConv support for HeteroGraphConv (#383) (@rbSparky)
• Add SAGEConv support to HeteroGraphConv (#384) (@rbSparky)
• Error in docs (#385) (@rbSparky)
• Add GINConv support to HeteroGraphConv (#390) (@rbSparky)
• Add ResGatedGraphConv support to HeteroGraphConv (#391) (@rbSparky)
• Adapt 4 convolutions to TemporalSnapshotsGNNGraphs (#392) (@aurorarossi)
• Add more conv layers with TemporalSnapshotsGNNGraphs support (#393) (@rbSparky)
• Added Usage Examples for Graph Convolutional Layers (#396) (@rbSparky)
• Add example for gender classification on TemporalBrains dataset (#397) (@aurorarossi)
• Add GATConv support for HeteroGraphConv (#400) (@rbSparky)
• New add_self_loops(g) for hetero graphs (#402) (@rbSparky)
• Updating the gnngraphdocs.md (#403) (@achiverram28)

Closed issues:

• implement add_self_loops(g, edge_t) for heterographs (#329)
• add empty constructor for heterograph (#338)
• getgraph not working on GPU (#377)
• Example given for GNNGraph results in error (#380)
• Error in one of the examples in the Working with GNNGraph page (#401)

v0.6.16

GraphNeuralNetworks v0.6.16

Diff since v0.6.15

Merged pull requests:

• Add heterogeneous add_self_loop support (#345) (@AarSeBail)
• CompatHelper: bump compat for Adapt to 4, (keep existing compat) (#354) (@github-actions[bot])
• Custom Norm Func for GCNConv (#356) (@rbSparky)

Closed issues:

• GCNConv without normalization (#277)

v0.6.15

GraphNeuralNetworks v0.6.15

Diff since v0.6.14

Merged pull requests:

• example correction (#347) (@eahenle)
• CompatHelper: add new compat entry for Statistics at version 1, (keep existing compat) (#350) (@github-actions[bot])
• Set compatibilities for standard packages (#352) (@bicycle1885)
• Relax CUDA compatibility (#353) (@bicycle1885)

Closed issues:

• Error in Example Code (#346)
• Outdated default package installation (#348)
• View arrays on GPU cause scalar indexing error (#349)

v0.6.14

GraphNeuralNetworks v0.6.14

Diff since v0.6.13

Merged pull requests:

• batch wider eltype (#340) (@CarloLucibello)
• Remove type constraints on Flux.batch for GNNHeteroGraph (#342) (@AarSeBail)

Closed issues:

• The constraint in Flux.batch(gs::AbstractVector{<:GNNHeteroGraph}) does not seem to be strong enough (#341)

v0.6.13

GraphNeuralNetworks v0.6.13

Diff since v0.6.12

Closed issues:

• Edge weights not properly documented for GNNHeteroGraphs (and implement new function to add new edge weights?) (#331)

Merged pull requests:

• Bump actions/checkout from 3 to 4 (#336) (@dependabot[bot])
• improve add_edges for graph and heterographs (#337) (@CarloLucibello)
• extend reduce_nodes and graph_indicator for heterographs (#339) (@CarloLucibello)

v0.6.12

GraphNeuralNetworks v0.6.12

Diff since v0.6.11

Closed issues:

• Update Flux.trainable (#323)
• HeteroGraphConv bug: ERROR: duplicate field name in NamedTuple: "movie" is not unique (#332)
• add_edges for GNNHeteroGraph does not allow providing the number of nodes (#334)

Merged pull requests:

• Update Flux.trainable (#330) (@aurorarossi)
• Fix heteroconv aggregation + tests (#333) (@svilupp)
• fixes add_edges + adds tests (#335) (@svilupp)