Planning HPC Python at HPC2N and UPPMAX

Materials pages

  • material: https://uppmax.github.io/HPC-python/index.html

  • source: https://github.com/UPPMAX/HPC-python/tree/main/docs

Dates spring 2026

None decided at the time of writing

See the course schedule

Course dates:

  • 2026-04-20

  • 2026-04-22

  • 2026-04-23

  • 2026-04-24

Meeting 2026-06-12 10:30-11:30

Present:

  • Birgitte

  • Björn

  • Jayant

  • Pedro

  • Richel

Goals

  • Talk about evaluations

  • Next iteration

Evaluations

General: large spread in confidence: - Always some that have high confidence and at least 1 that jave no idea even though attending - Discuss Goal: Have more “High” or less “low” or do we have to have “both”? - RB: I think a teacher may aim for ‘high’, but as far as I have searched the literature, I could not find a relation between this value and course quality, hence I ignore this value completely. I’ve only informally observed it to correlate to the day of the course (it decreases per day). The only value I see in self-assessed confidence is that there is less literature against it: course satisfaction has been more strongly proven to be useless to assess couse quality.

  • Maybe, if our learners are new to Python, only follow Day 1?

  • We are discussing different topics during the day. Without knowing everything, can they finish the course? I think that will be challenging

  • Maybe we should make it clear what the level is of the different parts?

  • Let us clarify what the level is of the different topics!

Day 1

Day 2

  • Day 2 summary

  • Day 2 full evaluation

    • Least confidence: virtual envs, IDEs, interactive, install

      • RB: as far as I’ve read about evaluations by learners, there is no use to compare the results (i.e. confidences) between sessions or teachers: only a teacher teaching the same session multiple times may think it is useful to see this number change

  • Most problems with Dardel users

  • The afternoon was split up differently this time

Day 3

  • Day 3 summary

  • Day 3 full evaluation

    • Least confidence: Formats Big data, Seaborn, Work inside Pandas

  • Most problems with Dardel users

  • We can point more to the NBIS courses on analysing big data

  • We can point more to other courses in general

Day 4

  • Day 4 summary

  • Day 4 full evaluation

    • Least confidence: parallelization, GPU, Tensorflow

  • Maybe some learners will only join on Day 4, hence they may not have installed an ML virtual environment as done on Day 3. We adapt by the Day 4 courses by starting with a ‘do this’ (i.e. create the virtual env), then talk, then use the virtual env

  • Should we add an Intermediate or Advanced Python Day? It is a lot of work and we expect to few learners to make it worth it. We seem to be leaning to a no. Instead, we could point more to other advanced courses

  • We consider to make more connected exercises here

  • We will make it more clear that they can get a course certificate for the days they attended

Next iteration

  • Keep day 1 outside or keep as refresher?

    • Refresher/prerequisite

  • Should we extend the course by 1 more day? It is often mentioned in the evaluations. Some of us think we definitely should do so. - [DECIDED] Yes: BB, PO, JY - Neutral: BC - No: RB

  • If we extend the course, we can schedule these days over more weeks

    • One week: RB, BB

    • [DECIDED] Two weeks: BC (), BB, JY, RB, PO

    • Three weeks:

  • There are many problems with Dardel. Can we use Arrhenius and if yes, should we focus om it?

    • Arrhenius will be main course project

      • Tetralith: will be decomissioned

      • Alvis: decomissioned or local (the hardware will be reused someplace or sold: we don’t know :-)

      • Dardel up until Dec (at least)

  • Point more to other courses

  • , prereqs etc

  • Follow up in August/September,

  • Should we give this course in autumn yes/no?

    • Yes: BC, RB, BB

    • Neutral:

    • No:

  • How often should we give this course? Maybe once a year, yet, due to the new HPC cluster Arrhenius, we should have an extra course to help these new Arrhenius uses

  • Should we transition to:

    • NAISS GitLab

      • Yes: BC

      • Yes, but I would not do so if I would be boss: RB

      • Neutral

      • No

    • Zensical

      • Yes: BC, RB

      • Neutral:

      • No:

  • When should the next course be appoximately?

Meeting 2026-02-26 14.00

Present:

  • Birgitte

  • Björn

  • Jayant

  • Pedro

  • Richel

  • Sahar

  • Rebecca? Otherwise Sahar / Richel

Goals

  • Get on track

  • Go through evaluation

  • Identify needed changes (if time)

  • Available teachers

  • Advertise and registrations

Agenda

  • Teachers available (a list can be found below)

  • Evaluations

  • Changes

    • Sphinx –> Zensical

    • GitHub –> Gitlab?

      • Can learners make issues?

      • Would it work if the repo is private for us?

        • No! We need it for, e.g. downloading tarballs, allowing learners to submit a PR, etc.

    • big picture

  • Advertise and registrations

Teachers available

  • 2026-04-20: Day 1: Richel

  • 2026-04-22: Day 2: Richel’s sessions go to Björn, Björn, Birgitte, Jayant, (Sahar)

  • 2026-04-23: Day 3: ?Rebecca, Sahar as backup for Rebecca’s sessions, Björn, Birgitte, Sahar

  • 2026-04-24: Day 4: Birgitte, Jayant, Pedro

Misc:

  • Birgitte will ask Joachim for status of Rebecca at the coordination meeting

Evaluations

Overall

  • Very few answered: 4-5

  • want more about COSMOS-SENS because little documentation at LUNARC

  • practical management of Conda/virtual environments on HPC: we can attempt to put it back in

  • Workflow systems like NextFlow or Snakemake: this is other courses

  • I guess that Python, pandas, matplotlib and such could be prerequisites to the course and instead only focus on big data and HPC specifics

    • The order is right: Day 1 teaches Python needed for Day 2, the Day 3 morning teaches the prerequisites for its afternoon

    • We keep it as it is!

  • The material is good, but the teaching technique is lacking.

    • Richel started off very good with a lot of interaction, breakout rooms and time for questions and doing things yourself.

      • Teachers decide on how to teach, as per usual

    • I liked this and that the exercises were part of the teaching, not meant to do during breaks, then you would never get breaks

      • There is the intention to add more time for exercises

Day 1

  • Day 1 had 3 responses (38% fill-in rate)

  • Success score: 90%

  • Comments:

    • It’s very nice to be able to through everything at your own pace, and still receive help/feedback on where you are, independent of the rest of the group.

    • The instructor was so helpful and friendly and helped me a lot in learning new stuff and solving issues I had during the day

    • Great teaching!

    • Great work with creating engagement

Day 2

  • Numbers answering: 4

  • Success score: 82%

  • Best confidence: Find & load / Install Python packages

  • Lowest: Check if in interactive session

    • This session is low on time (as is also mentioned by the learners)

    • There needs to be taught too much different ways

    • May fix itself when we have Arrhenius in the future

    • Working group decides on how to split the time up better and will schedule a meeting: BC, BB, JY, (RP)

  • Comments:

    • Exercises were a bit less clear today, which ones we should do and when was a little confusing.

    • Because each HPC has different setup it is a bit uninteresting to walk through them all when I’m only concerned with the one I’m using

      • We feel we are doing this more already

    • a little bit tight

    • The time allocated isolated environments and launching IDEs from the command line did not seem adequate

    • I and some other seemed to be lost already at OnDemand, and interactive work.

      • So maybe you should make that session more interactive. Then it would have been easier to follow the other sessions after that.

Day 3

  • Numbers answering: 5

  • Success score: 66%

  • Best confidence: Create plots in matplotlib

  • Lowest: Create plots in Seaborn

    • May be a leftover from incomplete discussion on IDEs

  • Comments:

    • when you get stuck in setting things up you then also loose on the learning

    • Overall a good setup of the course! It is hard to do the exercises because environments etc. has to be set, so I will have to go back later to finish.

    • Demonstrating everything with actual code and a more hands-on approach would be essential to better understand the core concepts.

Day 4

  • Numbers answering: 4

  • Success score: 52%

  • Best confidence: Multiprocessing

  • Lowest: Determine which ML/DL modules are installed at centre

    • BC will try to make this better

  • We will have InfraViz around again next time

  • Comments

    • a bit too fast for me to really learn by following examples, too uncertain with setting up environments, connecting etc

      • People jump in on only Day 4, where the prerequisites (e.g. connecting) are taught earlier

      • Suggest: for each day, have a link to the requirements of that day

      • Maybe connection session in the morning? No, we tried this and nobody showed up.

    • more on basic usage with nodes, set up Jupyter, environments, loading modules

      • We assume he/she did not follow the earlier days …

Changes

  • Sphinx –> Zensical

    • Let’s wait until the other courses have moved on, as there is also a GitHub –> GitLab move

  • GitHub –> GitLab

    • Let’s wait until the other courses have moved on, as there is also a Sphinx –> Zensical move

  • Big picture changes in lessons?

Notes from previous meeting