2025-04-25

My session went reasonably well. I felt tight on time, and there were 2 technical problems, but success rates were high (see below) anyways.

I’ve seen that a colleague discussed how to get the tarball in her session. Maybe it should be moved to her session? Nah, it fit fine within mine. Instead, just remind.

  • Remind colleague that I do the tarball exercise

There was a technical problem: my computer partially froze while discussing the exercise. Learners said they could still hear me. As I was nearly done talking, I just said them to work from this clumsy setting. I gave them 5 minute to work individually on the exercise. During those 5 minutes, I logged out and logged in again. All in time! I was happy with this improvisation.

Due to this, there was another technical problem: my rights as host were lost. This meant that I could not create breakout rooms. I improvised: I gave the learner and end time, up until they could do anything they wanted: I even encouraged them to take off their headset, as (1) I want them to work undisturbed, and (2) I would take questions in the main room. This worked out reasonably well. I was happy again with my improvisation, although it would have been more fun/useful for the learners to have worked in groups: it also made me more blind of their progress … but, hey, it was the best I could do :-)

In the end, I did ask the progress of the learners:

  • 13 out of 13 were able to use a Python module

    • This 1 learner struggled with logging in to Dardel

  • 12 out of 14 installed the tarball

    • 1 is the 1 learner above

    • For the other learner is unknown why this did not work. I forgot to make a note who this was

Regardless of the hassle, most learners were able to complete the exercises.

I think the camera recordings do more harm than good, as it kills interactivity during class, at the minimal gain of people outside of class being able to see these recording:

  • Using live recordings as a supplementary tool may not be worth it in our settings, as ‘No performance advantage was found for viewing recordings despite students’ positive perception of their utility.’ [Lesser, Melissa, and Lora Packel. “Educational Lessons from a Pandemic: Lecture Recording and Learning Outcomes.” Internet Journal of Allied Health Sciences and Practice 22.2 (2024): 9., Link to paper]

  • ‘The main finding is that students using the online lectures as a substitute for attending lectures are ultimately at a fairly severe disadvantage in terms of their final marks’ [Williams, Andrew, Elisa Birch, and Phil Hancock. “The impact of online lecture recordings on student performance.” Australasian Journal of Educational Technology 28.2 (2012), Link to paper].

  • I found most experimental papers do not compare with live recording (it is usual pre-recorded versus an unrecorded/live lesson). The exception is a paper that does recommend the recording live webinars, although it admits that ‘[Webinars tend] to be less engaging and less interactive for the participant’ [Dailey-Hebert, Amber. “Maximizing interactivity in online learning: Moving beyond discussion boards.” Journal of Educators Online 15.3 (2018): n3., paper].

  • Suggest this in a meeting

For the course as a whole, think following the pedagogic literature has a bigger impact over reading evaluations.

  • For example ‘long lectures and presentations will fail because students stop viewing and listening after about six minutes’ [Nilson, Linda B., and Ludwika A. Goodson. Online teaching at its best: Merging instructional design with teaching and learning research. John Wiley & Sons, 2021]. From the same book, the 25 principles of learning from cognitive science (chapter 4, page 92) gives some great pointers, e.g. ‘2. Students learn […] better […] by engaging in an activity than when they passively watch […] an instructor’, ‘6. Some qualities […] help students learn […] better […]: human faces, […] instructor enthusiasm’.

  • Suggest this in a meeting

Overview of how time is spent

Hour

Lecture (mins)

Dialogue (mins)

Exercises (mins)

1

15

5

30

2

35

4

11

3

38

5

0

3.75

17

0

0

4

49

11

0

5

54

2

0

6

0

2

28

Pace

Pace:

  • the pacing was good. I really liked the course material.

  • Fast on the second day

I assume this is not about my session, as 60% of its time there was time for exercises.

  • great

  • I think it was good, I could keep up and take notes during all of the lectures.

  • I feel like it was a bit rushed, good content but a bit too much given the amount of time maybe?

I assume this is not about my session, as 60% of its time there was time for exercises.

  • It was a lot of information to take in. However the resources provided are sufficient to catch up if we were stuck anywhere.

  • Good

  • It was the right pace

  • I did not feel comfortable during the afternoon session was running too fast, just to adjust to the schedule and got few from that and was confused because of several concepts in a bunch without explain the practical examples I guess it is been assumed that everybody knows all the concepts.

Not my session.

  • Too fast to combine explanations and exercises

I assume this is not about my session, as 60% of its time there was time for exercises.

  • Good pace considering the amount of materials and time. Then one needs to practise home by rewatching the YT videos

Hmmm, in general, I find that live recordings have ‘No performance advantage was found for viewing recordings despite students’ positive perception of their utility.’ [Lesser, Melissa, and Lora Packel. “Educational Lessons from a Pandemic: Lecture Recording and Learning Outcomes.” Internet Journal of Allied Health Sciences and Practice 22.2 (2024): 9., Link to paper]. However, that paper does find positive effects for learners that, like here, use it after attending a session.

  • Inconsistent - seemed slow much of the time, while some parts passed by before they could be properly explained or understood: it seemed to me that the time was not spent on the right things. I know enough about Python, I do not know enough about the software module system and SLURM and any other HPC-specific stuff.

I assume this is not about my session, as 60% of its time there was time for exercises.

However, I will suggest to do more foundational things at Day 2

  • Suggest to do more foundational things at Day 2 in a meeting

  • Too fast

I assume this is not about my session, as 60% of its time there was time for exercises.

  • okish

  • Suitable

Future topics

Future topics:

  • Xarray and dask

  • it is totally enough for me to study this content, and the document is great, I can read the document to revise the lesson

  • It could be helpful to have a lecture about R as well.

I don’t think R should be part of a Python course. But hey, I can link to it

  • Link to R course

  • More information on the HPC environment within a similar context. Maybe focusing on one HPC centre at a time.

  • Topics are ok. Just give more profound and practical demos during the sessions.

I assume this is not about my session, as 60% of its time there was time for exercises.

  • A summary of options in using Python in the HPC, with suggestions and best practices. What to do when the default modules are not enough.

The ‘Install packages’ session should cover that.

  • This 2nd day and we have not going through ML and Deep learning PyTorch and Tensorflow GPU: PINNs, Quantum ML (PennyLane, Qiskit) or small LLM

Sure, we do that on Day 4.

  • matplotlib from a regular shell/interactive/in compute jobs

  • How to find compatible modules and packages

We could add this indeed.

  • Suggest this in a meeting

Other comments

Other comments:

  • every thing is great for me

Great!

  • I think all of the training materials were very clear, and I liked the amount of hands-on exercises. I also really appreciated how supportive all of the teachers were, the classes had a very nice atmosphere.

Great!

  • Overall, a great session, but it would be great to have more time planned into each session for questions.

I assume this is not about my session, as 60% of its time there was time for exercises and questions.

  • For the matplotlib part: to improve more demonstrations rather than read so quickly the content to accomplish the schedule. Improve the depth of the course not too superficial to at least catch some key points that helps to start. The morning session had better demo and I could follow the steps.

Great!

  • The written material is quite good and shows the different options, but simply going through them during the course takes too long and sometimes end up with focus on specific HPC systems that not everyone is going to use. There are instructions for each cluster, so maybe the topics should be more generic with access to the specific info as needed. The IDE section, for example, could only show how to do it to very specific cases, while having the cluster instructions and a generic discussion about the options could be better.

I assume the ‘focus on specific HPC systems that not everyone is going to use’ does not apply to me: I directly start from the cluster-specific documentation.

  • Overall very good docs from each mentors that states pro and contra and niche details of each HPC. Infinite for trouble-shooting support especially for beginners

Great!

  • (did not attend Matplotlib part) 1. Overall the reading/presentation material was good to very good, just too far between exercises - as soon as one step has been explained, I want to test it and make sure I can do it! But some information was outdated and some part or parts had lots of spelling errors. 2. Again and again, there was not enough time margins within each session. Asking us to do stuff on the breaks is BAD. Breaks are needed for food, toilet and looking away from the screen. Can’t learn anything if we don’t get breaks. 3. Lunch break start and end time needs to be confirmed before 12:00. For those of us caught up in listening and following along it got cut much too short today.

Fun that the spelling errors are mentioned!

  • Mentioned this in a meeting

Indeed, the breaks were not respected.

  • Mentioned this in a meeting

  • It’s good that you record, but could you post the recordings a bit faster so we can go through before the rest of the course?

I assume this does not apply to me, as my recordings are in my course material.

  • Not fond of the IDEs, I think it’s unnecessarily cumbersome. Shell is much easier to work in. You don’t have to load a lot of other things, set graphics and fight with it :( Why this focus on jupyter or spyder or whatever it was called

I agree that not everyone needs an IDE. However, Jupyter was needed to be able to do the exercises by Anders Hast.

Detailed feedback from a learner

Detailed feedback from a learner:

NB: On the first day, with Richel, I was mostly in the role of a TA rather than a learner. I did not fill out the feedback form for that day, but I had an immediate evaluation meeting with Richel at the end of the day.

…

NB 2: I did fill out the feedback evaluation form for the three remaining days, and did not look back on what I wrote then (or look at the course evaluation results at all) while writing this. Therefore this document probably duplicates some of that while being feedback from a single (and perhaps not the most representative) learner.

…

NB 3: I did not attend the matplotlib sessions.

A multitude of teachers

This is GREAT! Some variation in perspective, focus and intensity makes a big difference and is much appreciated. Anders Hast deserves special mention for bringing true researcher enthusiasm, it is great to learn from teachers that know and enjoy the subject at hand like that.

I agree: I liked Anders’ lecture too! Even though there were 6 minutes of exercises (on the YouTube video: 28:47-34:02), I do think this is worth it.

  • Mention this in a meeting

Sometimes a useful connection between two different sessions gets more or less lost, e.g. mentioning line-by-line file IO from the first day when talking about file formats at the end of day 3 would be easier to do well in a traditional one-teacher-per-course format. For this course I deem these losses very small and acceptable, especially in the light of the advantages mentioned above.

I think this will be solved when more exercises are done.

However, the contrast between sessions pulls into sharp focus a few things to improve upon:

…

Preparedness is key

…

Several sessions had examples, code-alongs or exercises that did not work as intended without troubleshooting that was not within the focus of the session. Learning about modules and packages and handling the dependency hell in various ways is very good (I’ll get back to this), but it is not good to be interrupted by these things when trying to teach and learn something else.

I agree.

  • Mention this in a meeting

It would be good if all of the course materials, for all of the clusters, was confirmed working as intended before the course.[1] I do realize that this is a big undertaking—but make no mistake, this course (with the amount of content it currently has) is definitely a big undertaking.

I think I already do this, as I’ve checked my course material by actually testing this by recording.

Too much content, too little time

…

Full days leave little to zero time for going through the exercises on our own. Having extra exercises for repeating and/or diving deeper after the course is good, but all that is truly intended to be part of the course needs to be given the time within the course. With no time to test and get some confidence that one part is understood and doable, open questions and uncertainty cause several problems for learners.

I assume this is not about my session, as 60% of its time there was time for exercises and questions.

First, it immediately steals away some focus/brainpower. Second, we might not get knowledge/skill presumed by later parts of the course. Third, the sheer worry of the latter is not to be trifled with! I have seen that wreak havoc on learning again and again in various lectures and labs.

I agree that we should test if our learners got it.

  • Mention this in a meeting

The sessions on Pandas and Seaborn especially were much too rushed, to the point that they felt close to useless. The session on IDEs had barely enough time for me to get Jupyter running, while I would’ve expected to try VS Code and Spyder as well since all of them are in there. The course taken as a whole also needs either less content or more time.

  • Mention this in a meeting

To solve this nicely, please consider what content is crucial that learners get to try and to ask questions about! The rest can be left in extras for the learners to look at after the course. This is closely related to how general a topic is, if there are many other places to ask questions and get the right answers.

I hope I do this

We learn what we do

…

In some sessions, mainly Parallelism and ML/DL if I recall correctly, I was learning about modules and packages and lost track of the lecture as I was trying to make the examples work. In some sessions I learned little to nothing as I found no time to try anything. In some sessions I got to try and learn every single thing that was in the course materials.

I hope the latter applies to my sessions, as there was enough time for exercises.

There seems to be two good ways to have learners actually do stuff in an online course. One is what Richel does: most of the time in 2-person breakout rooms, working through exercises as the main way of learning.

Hey, that is me!

Another one is to have a well-prepared code notebook with demos/examples/exercises to be completed by the learners. On desktop screens and larger laptops learners can have their copy of the notebook side-by-side with the meeting. This method can smoothly intersperse both longer and very short bits of hands-on coding into a lecture, akin to how the demos that weren’t working in the sessions on Parallel programming should have worked.

Focus and structure

This course teaches many separate things. This gets a bit too messy, both for the learners who do need to learn all of it—these people are new to the entirety of this and will not be able to group and separate things—and for the learners who only needs selected parts of the course.

…

It would be easier for both groups (though a bit less efficient, for those who both need all of it and are somehow able to efficiently learn it the way it is now) if this content was divided between as much as four separate courses:

…

  • Intro to HPC, usage and management of software and code (Logging in, LMOD, SLURM; the things that are actually different between HPC and a single-user single computer)

  • Intro to Python (this already exists and is day 1)

  • Python coding for big computation, big data and AI (this includes parallell and GPU but is not specific to HPC and probably exists elsewhere)

  • Using Python code, packages and environments at NAISS HPC systems (LMOD, venv, conda; when to use what and how they interact)

…

Course 1 is general to HPC users, not specific to Python. It should probably be provided by NAISS quite often and be a prerequisite to all other NAISS courses. Courses 2 and 3 are general to Python users, not specific to HPC, widely available from elsewhere. Course 4 is specific to Python in HPC; this is what I, a seasoned learner, would expect NAISS to provide in regards to using Python.

I enjoy this suggested schedule. I know that ‘Intro to Python’ will get more HPCey, among others by adding how to loading Python package modules. I feel that is a first step to what the learner wants/needs.

What I expected (or hoped) to learn that I did not learn

Essentially the fourth course mentioned above. I have not acquired the one skill (AFAIK) that is truly specific to Python on HPC: how to choose how to manage the packages. Everything else (that was in this course) is shared with either other HPC users or other Python users—therefore it is discussed and taught by other resources—so this is what I truly wanted from this course. I did learn a bit about it but it seemed to happen by accident as much or more than as a result of the course truly focussing on this matter.

And:

The fix would be to make sure every single command that loads a module or a python package does specify a specific version. Well, for the recurring problem that I do remember, but there may have been more problems.

As far as I know, my course material has explicit versions.

  • Check that my course material has explicit versions

Some other feedback

From Day 3:

This is mostly for the parts last week, I’m not so fond of this calling out of people with their names. It is stressing. I’m here to learn, and I do that better in peace.

I assume this applies to me, and maybe to other colleagues as well. It is too bad that 16.1% (in range 6.1-25%) of the Swedish population, has social anxiety, although ‘Subjects with social phobia who do find their way into the university could be described as having predominantly a mild form of the disorder’ [Tillfors, Maria, and Tomas Furmark. “Social phobia in Swedish university students: prevalence, subgroups and avoidant behavior.” Social psychiatry and psychiatric epidemiology 42 (2007): 79-86.] Source. The effect of this is unknown, as here I quote the paper: ‘future studies should investigate whether sufferers of social phobia underachieve or abolish their studies prematurely.’. The main solutions given in that paper is ‘It is an important task to inform students with social phobia that effective pharmacological and psychosocial treatments are available.’

I will keep doing this, as asking people randomly is most effective (where raising hands is least effective) [Bell, Mike. The fundamentals of teaching: A five-step model to put the research evidence into practice. Routledge, 2020, pages 128-129]. However, I do introduce that I will ask people randomly, giving people with social anxiety a way out (by turning off the camera) and demonstrating that an incorrect answer or a ‘I don’t know’ is fine too. So, I agree with the learners, yet his/her solution is not the one I will follow.

Also, could you please post the recordings a bit faster next time. I saw in the Q/A document that Brigitta lost the recordings for the first day so I am happy that Richel at least posted them! Would be better to have them in the same youtube list as the rest though

I think this is a bit harsh on Brigitte. Instead of uploading the video twice, it can be linked to from different playlists. Alternatively, what about we put the playlist on the website?

  • Suggest to do so in a meeting

Meeting notes

Below are the meeting notes of 2025-05-22 with my suggestions.