Community call - 2024.02.29

[gdalle]

Tentative agenda

• Benchmarking efforts - @filchristou
• Restructuring organization repos (flows, matchings, optim, community detection, extras) - @gdalle @simonschoelly
• Graphs.jl submodules (Experimental, Parallel, possibly Contrib) - @gdalle @simonschoelly
• New graph generators? - @fcdimitr

[gdalle]

Benchmarking

• Different approaches possible
• Basic functionality: BenchmarkTools.jl and PkgBenchmark.jl
• Integration with CI: BenchmarkCI.jl (unmaintained)
• How to store and display benchmark results? Ideally in the PR comments
• Problem: PRs from public forks don't have write permissions to the target repo (for security reasons)
• Solution: chained workflows?
  1. One which runs code from a fork (potential risk), stores the benchmark results as JSON in the workflow uploads
  2. One which takes the workflow JSON uploads, formats them and puts a comment in the PR
• No need to keep benchmarks from long ago
• ITensors.jl uses a similar workflow
• For testing, @filchristou needs to add a workflow file to master and experiment with it, vote with reacs

[gdalle]

Graph generators

• Dimitrios was developing packages for community detection
• Need for new graph generators that are used as benchmark: Lancichinetti-Fortunato-Radicchi
• Pull request involved external dependencies with C++ author code: Lancichinetti-Fortunato-Radicchi generators Graphs.jl#345
• Wrapped in a jll in https://github.com/DiNiXi/lfr-benchmark-c and added to Yggdrasil
• Basic Graphs.jl should have very few dependencies
• Solution: do a Julia rewrite and then submit it to the main package
• Don't put the C++-dependent code in GraphsExtras, put it in a personal repo
• Same idea with https://github.com/fcdimitr/fglt, relies on linear algebra
• SuiteSparseGraphBLAS could be an extension cause it is generic, but not algorithm-specific binaries

[gdalle]

Other organization packages

Community detection:

• Empty and archive CommunityDetection.jl, move code to Graphs.jl
• Remove kmeans dependency for spectral methods, just compute the embeddings cause the user is free to cluster however they want, or just visualize
• Separate page about "community structures" into metrics and algorithms

Flows and matchings:

• PR underway in Graphs.jl: Migrate flow algorithms from GraphsFlows Graphs.jl#329
• First @etiennedeg wants to merge fix floating point GraphsFlows.jl#3
• Put it in Experimental? GraphsFlows.jl is stable at this point
• Ensure GitHub clarity by doing first a copy-paste and then changes

Can we put new algorithms in the main namespace if the PR is readable? Vote with reacs

[gdalle]

Submodules

• At the moment, Experimental, Parallel, LinAlg, SimpleGraphs, Test
• More thematic submodules could help resolve namespace clashes
• But this can also be solved by dispatch or name changes
• Why is there a SimpleGraphs submodule?
• The discussion is a long term one

[gdalle]

Graphs 2.0

• No progress
• Get funding? @gdalle will be a PI soon, can start looking for grants