Meeting 17 Oct

Agenda

• Action points from last meeting

• Schedule

• Admin

Action points (DECIDED in May)

• More InfraVis-like collaborations demoing usecases

  • When? ML afternoon 2 Dec

  • Data visualization

  • This iteration? Yes

  • Who does? Jayant

• Move ‘Slurm and batch jobs’ from Day 3 to Day 2:

  • BC: We already had the introduction?

  • OK toatally second day

• Pinning versions on the clusters (already doing it partially)

• prereqs

  • courses

    • login/transfer

    • modules

    • slurm: keep condensed session focus on Python

• Will will run Markdown next time

  • Still yes??

  • BC: MkDocs should be used then directly

    • 2-step is too expensive (md with sphinx)

  • Delay to May: BC, PO, RP, JY, BB

• Times

  • Have 15m always

  • Lunch 60 min

• Give Anders Hast more time

  • In ML session: YES

• DECIDED Mark exercises as important and optional: RB, BB, BC

• IDEs: keep the schedule as it is

  • DECIDED Present all 3 IDEs but parallel exercise tracks (where learners decide themselves): BC, RP,

• Videos

  • DECIDED Youtube playlist + table on course web (also NAISS web)

  • the latest available (living table)

Not decided because postponed

• Recordings

  • DECIDED: always before or during

• Exercise rooms (perhaps later decision)

  • Let teachers decide how they teach: RB

  • always try to keep 2-3 persons in each room (BC and RB has good experience here): RB

  • larger rooms: RB

  • cluster rooms: RB

  • combination of above: RB

• [Richel] One fun thing we did in another course, is to let the learners rename themselves, to include the cluster name, e.g. Thora becomes [Dardel] Thora

  • Let teachers decide how they teach: RB

  • Let’s do this in the introduction: RB

• [Richel] In some courses we measure the initial confidence our learners have at the start, such as Day 1. What about we do this at the other days too?

  • Never measure the initial confidence: RB, RP

  • Keep it as it is: RB

  • DECIDED Let’s try to measure the initial confidence at eachday: RB, BC, BB

  • Need to formulate well in such case

• BB: Interconnected courses: we need to connect our courses better

  • Use a working group (i.e. one working group per workshop): BC, RB

  • Agree: RB, BB

  • Not yet:

  • No:

  • Just make sure students know which courses should be taken in what order - need to decide on “curriculum”?

• JY: ML/DL should be more about workflows than core principles. Core principles should be its own separate course.

  • Yes (next time): BC

  • no

• JY: Do HPC centeres have the packages by Google Collabs?

  • yes:

  • no: To be discusses in future

• JY: Recording professionalisms, there are better tools than Zoom

  • Postpone

• JY: How to do recordings, i.e. blurring out names?

  • BB: knows the GDPR rules

  • Postpone

Schedule

Input

• less of matplotlib, pandas, seaborn

  • DECIDED keep some: BC

    • (good to understand how to make graphics avail and as prereq to big data)? BC

    • matplot/pandas 45-50 m x 1

      • load+examples+what they can get out of it + workflows

      • before big data; mention ML prep, memory-saving datatypes

  • VOTE remove all? RP (overruled)

Fri

• Theme: packages and IDEs + basic HPC including batch jobs and allocating GPUs?

• load-run (skip or shorten because included in first day?)

• longer with packages (perhaps input from pavlin about pixi and uv)

• intro to slurm and interactive (only)

• IDEs

• we are lacking analysis: should we remove all YES

Mon

• Theme: Running and analysing

• GPU (intro + Python specific)

• analyzing: short versions of earlier (matplotlib, pandas, seaborn)

• big data (dask)

• we are lacking analysis: should we remove all?

Tue

• Theme: ML and development?

• parallelism in morning 2 hrs

• ML

Todos

• Björn renews schedule

• Further details discussed in chat

Notes from previous meeting