GNNlib-v1.2.0

Diff since GNNlib-v1.1.0

Merged pull requests:

• feat: Reduce nodes over heterogeneous graphs (#634) (@lenianiva)
• fix: Empty edge handling in softmax_edge_neighbors (#636) (@lenianiva)

Closed issues:

• AMDGPU tests are failing (#632)

GraphNeuralNetworks-v1.1.0

Diff since GraphNeuralNetworks-v1.0.0

Merged pull requests:

• [GNNFlux] Translate Pluto notebooks to Literate (#569) (@aurorarossi)
• no constraints on numbers of features gdata for heterographs (#570) (@CarloLucibello)
• [GNNFlux] Translate Traffic prediction Pluto notebook to Literate (#572) (@aurorarossi)
• port GNNGraphs tests to TestItems.jl (#573) (@CarloLucibello)
• [GNNFlux] Fix Temporal graph classification tutorial (#575) (@aurorarossi)
• add reference to PyGDatasets (#578) (@CarloLucibello)
• fix GNNGraphs zygote error (#579) (@CarloLucibello)
• bump KrilovKit (#587) (@CarloLucibello)
• better inference for num_nodes (#588) (@CarloLucibello)
• fix corner case in normalize_graphdata (#589) (@CarloLucibello)
• Feature: TGCN should support non linear_activations (#596) (@juanmigutierrez)
• Provide CUDA support for the binarize() operation on sparse matrices (#601) (@dferre97)
• port tutorials from Literate to plain Documenter (#602) (@CarloLucibello)
• SpMM message passing CUDA support for coalesced COO graphs (#617) (@dferre97)
• fmt option in adjaceny matrix + propagate copy_xj for Metal (#619) (@CarloLucibello)

Closed issues:

• GraphNeuralNetworks.jl 1.0 release (#481)
• Can't create a GNNGraph over degenerate graph with just a single vertex (#580)
• GNNGraph default num_nodes to length of element of ndata (#581)
• Can't have empty edata for graph with no edges (#582)
• Generating and Feeding a Custom Dataset into T-GCN (#583)
• In Literate-generated tutorials the input code and the output have the same style (#590)
• GNNGraph constructor is type unstable (#593)
• Moving GNNGraph from GPU to CPU breaks Zygote gradient computation (#597)
• GNNChain doesn't work with layers that accept edge features (#598)
• UndefVarError: ones_like not defined in GNNLib in SGConv example (#599)
• add output format option to adjacency_matrix arguments (#606)
• add coalesce function (#608)

GNNlib-v1.1.0

Diff since GNNlib-v1.0.1

Merged pull requests:

• SpMM message passing CUDA support for coalesced COO graphs (#617) (@dferre97)
• fmt option in adjaceny matrix + propagate copy_xj for Metal (#619) (@CarloLucibello)

Closed issues:

GNNGraphs-v1.5.0

Diff since GNNGraphs-v1.4.2

Merged pull requests:

• Provide CUDA support for the binarize() operation on sparse matrices (#601) (@dferre97)

Closed issues:

• Generating and Feeding a Custom Dataset into T-GCN (#583)
• GNNGraph constructor is type unstable (#593)
• Moving GNNGraph from GPU to CPU breaks Zygote gradient computation (#597)
• add output format option to adjacency_matrix arguments (#606)
• add coalesce function (#608)

GNNGraphs-v1.4.2

Diff since GNNGraphs-v1.4.1

Merged pull requests:

• port GNNGraphs tests to TestItems.jl (#573) (@CarloLucibello)
• add reference to PyGDatasets (#578) (@CarloLucibello)
• fix GNNGraphs zygote error (#579) (@CarloLucibello)
• bump KrilovKit (#587) (@CarloLucibello)
• better inference for num_nodes (#588) (@CarloLucibello)
• fix corner case in normalize_graphdata (#589) (@CarloLucibello)

Closed issues:

• GNNGraph default num_nodes to length of element of ndata (#581)
• Can't have empty edata for graph with no edges (#582)

GNNlib-v1.0.1

Diff since GNNlib-v1.0.0

Merged pull requests:

Closed issues:

GNNGraphs-v1.4.1

Diff since GNNGraphs-v1.4.0

Merged pull requests:

• no constraints on numbers of features gdata for heterographs (#570) (@CarloLucibello)

Closed issues:

GraphNeuralNetworks-v1.0.0

Diff since GraphNeuralNetworks-v0.6.23

Breaking Changes:

• The temporal convolutional layers have been completely revised, both for adherence to the original papers (e.g. for the TGCN layer) and for compatibility with the new recurrence style in Flux v0.16

Merged pull requests:

• changes for Flux v0.15 (#550) (@CarloLucibello)
• rewrite recurrent temporal layers for Flux v0.16 (#560) (@CarloLucibello)
• broadcast, iterate, and setindex! for TemporalSnapshotsGNNGraph (#563) (@CarloLucibello)
• [GNNGraphs] fix show for non-array (#564) (@CarloLucibello)
• filter labels in diff for GNNGraphs releases (#565) (@CarloLucibello)
• filter labels in release diff (#566) (@CarloLucibello)

Closed issues:

• adapt to Flux v0.15 (#524)
• correctness temporalgraph implementation (#548)

GNNlib-v1.0.0

Diff since GNNlib-v0.2.5

Breaking Changes:

• modified set2set_pool and gated_graph_conv to adapt to Flux's v0.16 new recurrence style

Merged pull requests:

• filter labels in release diff (#566) (@CarloLucibello)

Closed issues:

GNNGraphs-v1.4.0

Diff since GNNGraphs-v1.3.1

Merged pull requests:

• broadcast, iterate, and setindex! for TemporalSnapshotsGNNGraph (#563) (@CarloLucibello)
• [GNNGraphs] fix show for non-array (#564) (@CarloLucibello)
• filter labels in diff for GNNGraphs releases (#565) (@CarloLucibello)

Closed issues: