Skip to content

2025-09-04 Technical Difficulties Round Table - Summary

  • No general sense on how to optimise Python code
  • "Don't use for loops" - but not really knowing what the alternative would be
    • numba decorators
    • Olivier: Has worked with numba before

  • Submitting batch jobs

    • How to estimate how much time it would take to run your code?
      • Just be conservative and overestimate (it would only be used in scheduling)
    • Not really clear how it works... Submitting jobs
    • VScode integration to SLURM is unclear
    • We can have a lunch talk on SLURM (and make sure that Michael K. is there so that he can hear our thoughts on SLURM and Lorenz).

  • JupyterHub on Lorenz?

    • Not really worth it since people can easily just use VScode

  • if you open up a folder on Lorenz, how can you open other folders?

    • Cmd+O: Open any file on your computer in vscode (regardless of in project)
    • Open multiple folders in project workspace: https://stackoverflow.com/a/45125674
  • Don't know how to use AI coding tools in vscode, is it paid?
    • GitHub Pro is available for all academics for free
    • Saves time copy pasting things. Sometimes its not always accurate.
  • More advanced git: How can we resolve merge conflicts?
    • There's a software carpentry course for git, with a section on collaboration (link).
  • Xarray and performance isn't really clear...
    • How to choose chunk size?
    • Why is dask (e.g., doing a spatial average for each file) slower than doing a for loop through each file?

Final takeaways:

  • VScode running in an interactive session would be useful on an IMAU level (and not just a Parcels level)
  • We should have a lunch talk on SLURM
  • An xarray talk series would be useful (also looping in performance discussions)