Community call - 2023.11.23

[gdalle]

Tentative agenda

• News from the ecosystem
• Benchmarking Graphs.jl before attempting big changes (@filchristou ?)
• A new interface checker (@olegfafurin ?)

[gdalle]

Benchmarking @filchristou

• Ongoing work in First try on benchmark CI Graphs.jl#315
• Taking a subset of the tests as benchmark? Very long and lots of code rewrite
• Multicore benchmarking not an option
• Benchmark basic access and modification of the SimpleGraphs
• Choose some algorithms based on existing benchmarks. Examples (maybe there's something better): https://qiskit.org/ecosystem/rustworkx/dev/benchmarks.html or https://www.timlrx.com/blog/benchmark-of-popular-graph-network-packages-v2
• Where to put the results of the benchmarks? Preferably in another repo
• Instead of uploading results elsewhere, put the markdown summary either in the PR or as a merge commit comment
• Would be nice to have the historical graphs: https://github.com/benchmark-action/github-action-benchmark/ does that
• Do we even need parallel benchmarks?
• Prettify the tune.json to diminish diffs

[gdalle]

DaggerGraphs @jpsamaroo

• Review the API: Add DaggerGraphs subpackage JuliaParallel/Dagger.jl#448
• Metadata partitioned like the graph
• Give a lazy accessor to avoid allocating the full metadata at once
• weights doesn't even have to be a matrix, overload getindex on your metadata object for instance
• Support for ingesting OpenStreetMap?

[gdalle]

Graphs.jl interface

• Maybe add metadata specification to Graphs.jl interface?
• https://github.com/gdalle/GraphsInterfaceChecker.jl
• Take inspiration from the tests in https://github.com/gdalle/GridGraphs.jl or DaggerGraphs
• Test API between two graph formats on the same graph object? Typically compare against a SimpleGraph