owned this note
owned this note
Published
Linked with GitHub
---
title: "2023 Summit: Meeting 1"
---
## Information
- Date: Monday, [February 27th 9AM - 10AM Pacific time (click for your timezone)](https://www.timeanddate.com/worldclock/converter.html?iso=20230227T170000&p1=224)
- Zoom Link: https://caltech.zoom.us/j/87686129450
- [Discord invite link](https://discord.com/invite/vur45CbwMz)
- [Summit topic planner](https://github.com/orgs/scientific-python/projects/4/views/1)
## Participants
- Jarrod Millman
- Stéfan van der Walt
- Brigitta Sipőcz (@bsipocz)
- Ross Barnowski (@rossbar)
- Juanita Gomez
- Inessa Pawson
- Tim Head (@betatim)
- Sebastian Berg (@seberg)
- Isaac Virshup
- Leah Wasser (pyOpenSci, she/her)
- Matthias Bussonnier discord:@mbussonn github:Carreau
- Sanket Verma - GitHub: [@msankeys963](https://github.com/msankeys963), Discord: `msankeys963#0660`
- Jim Pivarski [@jpivarski](https://github.com/jpivarski), Discord: `jpivarski#8588`
- Pey Lian Lim ([@pllim](https://github.com/pllim))
- Henry Schreiner
- Kira Evans @kne42
- Tyler Reddy
- Martin Fleischmann (@martinfleis)
- Kyle Sunden (@ksunden)
- Lars Grüter ([@lagru](https://github.com/lagru), D)
- CJ Carey
- Mridul Seth (@MridulS)
- Madicken Munk (@munkm on github)
- John Kirkham - [@jakirkham]( https://github.com/jakirkham )
- Levi John Wolf
- Julien Jerphanion [`@jjerphan`](https://github.com/jjerphan)
- Pamphile Roy (@tupui)
- Matt Haberland (@mdhaber, Matt H.#9697)
- Anderson Banihirwe
## Agenda
The summit is scheduled for 1 hour (hour -> week?) and will consist of a series of high-level
birds-of-a-feather (BoF)-style talks, followed by more focused discussion.
1. (5 min) Welcome & Introductions (Jarrod Millman)
Jarrod: Harder to have cross project, whole community dynamic, trying to make that better.
2. (10 min) Logistics
Stefan:
- Get everyone flights books via the UC travel agency, each will receive a booking code and contact UC travel agency. Exact date will be on the website.
- We'll try to use the money to cover next year, so if you have grant that would be great.
- Group accomodation later by Stefan.
- Feel free to reach out on Discord.
- Join here: https://discord.com/invite/vur45CbwMz
3. (30) Meeting Topics https://github.com/scientific-python/summit-2023/issues
- Jarrod Millman: Goal
- Unique opportunity to work on cross-project concerns
- We should think big, but focus on achievable short-term goals
build
- Stéfan van der Walt: Build systems
- Starting to see Meson emerge as an excellent build option for compiled Scientific Python libraries
- But there's no "standard" configuration/documentation for standard workflows yet; numpy/scipy/scikit-image are all feeling it out
- Editable installs just arrived, but work slightly differently to what we're used to with `pip install -e .`.
- [here is a basic pdm/meson build](https://github.com/pyOpenSci/examplePy/tree/main/example6_pdm_meson)
- Also starting to see more usage of [dev.py](https://github.com/scientific-python/devpy) as a convenient developer interface for various tasks
- Still in alpha; tool is quite general so takes some figuring out what commands should be provided / which flags to support
- https://github.com/scientific-python/devpy
- Leah Wasser & PyOpenSci is working on community guidelines for packaging
- One part of that is setting up small packages that use different tooling for comparison
- https://github.com/pyopensci/examplePy
- [newest pr for packaging guide here](https://github.com/pyOpenSci/python-package-guide/pull/55) <- feedback welcome!
- This may be a good place to capture some of the Meson workflows mentioned
- definitely and we plan to add tutorials too... so all in for talking about this (leah)
- There is therefore technical, user interface, and documentation work to be done
- [name=Henry Schreiner]: I'd like to see scikit-build-core included when looking at modern compiled build systems. It has simliar features to meson-python, and due to PEP 621, a lot of the basic setup is shared.
- [name=Henry Schreiner]: I'm interested in contributing the Scikit-HEP Developer guidelines (including cookiecutter with 12 build backends & WebAssembly Repo Review) - need discussion about what it would look like and would be called.
- Stéfan van der Walt: Sparse Arrays in SciPy
- In the previous release of SciPy we introduced experimental Sparse Arrays
- These arrays are limited to 2D, since they are built on top of Sparse Matrices
- We hope to refactor scipy.sparse:
- matrices should be built on top of arrays (if not deprecated entirely)
- sparse arrays should support 1D and, eventually, N-D
- sparse arrays should closely follows numpy array semantics
- [Isaac] Question: Could this include getting major packages (like scikit-learn) to support sparse arrays?
- [name=Julien Jerphanion] To some extends, it might. scikit-learn does support sparse arrays already but it has some rather niche needs. Those have been covered during [the first meeting of the Sparse Array Summit](https://scientific-python.org/summits/sparse/meeting1/) -- see [those slides](https://raw.githubusercontent.com/scientific-python/scientific-python.org-blobs/main/summits/sparse/meeting1/julien-jerphanion.pdf).
- [name=Julien Jerphanion] We do not (explicit) support the array api sparse arrays.
- [Isaac] Could this be task for this sprint? Getting support for these? I would love to work on this, since it's my main blocker for switching over the array apis.
- [Isaac]: Also I forgot from that meeting, but was using numba a possible solution for the scikit-learn sparse needs? I've done a fair amount of custom sparse ops in numba for scverse packages.
- [name=Julien Jerphanion] I do not think numba will solve scikit-learn's needs for sparse arrays (scikit-learn chose to depend on only a libraries, including numba).
- Brigitta Sipőcz: Shared infrastructure for SPECS and beyond
- SPECs: The current SPEC drafts are documents about policies the core libraries opt-in to follow (e.g. producing and using developer versions in testing), however, template implementation / suggested examples are needed to work out.
- help build examples of how to build nightly wheels and use them
- For example,
- https://scientific-python.org/specs/spec-0000/
- https://scientific-python.org/specs/spec-0004/
- https://scientific-python.org/specs/spec-0005/
- Testing: Libraries facing the same or very similar doctesting and docs building challenges, aggregating the needs and centralizing these efforts into a small number of pytest-plugins, and sphinx-extensions.
- isaac: myst + doctest? (I think that currently doesn't work)
- pllim: Would be nice if we can revive the pytest-doctestplus discussions. :)
- Tyler: using pytest + regex to run subset of tests (better version of `pytest -k "..."`)?
- Tutorial infrastructure:
- Build a template that enables easier maintaining of the various tutorial repos
- Is there missing features in the Jupyterbook infrastructure we would need?
- Matthias: I'm happy to give/move [my backport bot](https://meeseeksbox.github.io/) to Scientific-Python
- pllim: I love backport bot! It saves me and my colleagues so much time and it is so easy to install now.
- Matthias: That would allow SP to also "claim" all the things the bot has done.
- leah : wow matthias that looks awesome
- Matthias: I guess the bot to open issues when project drop support for older versions.
- Juanita Gomez: Community building & documentation
- Onboarding guides, content to help people get
- a guide for community managers in scientific python packages/communities
- [name=Sanket]: Sounds interesting and would love to contribute on this!
- How we collect metrics and package stats?
- Leah Wasser asked about that; Jim Pivarski is interested in joining that discussion, contributing some ideas about metrics.
- Inessa has been researching this topic and implementing some solutions in NumPy.
- [name=Lars Grüter]: Perhaps related, has there been interest in or work on something like (regular) user surveys? This might involve just shared infrastructure each project can use or even a shared "Scientific Python" survey.
- [name=Matthias]: WRT telemetry, there was a discussion/project 8(!) years ago at scipy: https://github.com/njsmith/sempervirens This was more how package can collect telemetry and how to have user "opt-in"
- Who will be at JupyterCon ? Can we bootstrap som of those things at JupyterCon sprints ? (may 13, 14, Paris, [jupytercon.com](https://www.jupytercon.com/))
- (maybe, visa stuff) Mridul Seth: We can probably do some work on the Jupyter tutorial infra? One pain point I have is that nbQA doesn’t work with myst notebooks, only ipynb. I really want to fix that soon.
- Matthias: You can ask for an invitation letter.
- Leah: Same question for pycon US! i'm happy to work on / talk about some things there as well
- Benchmarking (from Mridul Seth)
- Leah: there are some other topics we are working on around things like telemetry in open source that we could potentially talk about if there is interest given we are developing policy. it could just be a few short conversations for this type of topic. maybe the conversation relates around how to create policy with input from this community?
- [name=Martin]: I'd be interested in CI workflows that test my PR/RC against downstream packages depending on my package. Similarly to what CRAN is doing in the R world.
- [name=Sanket]: Release managers like Rever, Towncrier etc.
4. (10 min) Discussion
5. (5 min) Wrap up and next steps
## Notes