Ah yes. It is indeed the batchtools bug:

sbatch: [I] No *runtime* limit given, set to: 15.0 minutes
Submitted batch job 60933455

Fun!

I can see in the log that a job was started and failed after 3 s.

Omitting --jobs=runtime avoids the sacct error and yields: "FAILED"

Error: Failed to check if job is pending (exit code 1);
cmd: 'sacct --user=$USER --noheader --parsable2 --allocations --format=State --jobs=runtime --starttime=2025-09-09T22:37:52'
output:
sacct: fatal: Bad job/step specified: runtime

Rocky Linux 9.6 (Blue Onyx)
slurm 22.05.10
HPC info help.itc.rwth-aachen.de/service/rhr4...

I really like the look of this one!

Is the publication up on biorxiv yet?

I work mostly with multi channel fluorescent data, does it make sense to encode my favorite 3 channels as RGB and calculate features or is this more geared towards bright field?

You don't like the time honored ritual of torch.cuda.is_available() ?

Wow awesome! The only thing stopping me from using juv was the focus on Jupyterlab.

Cool paper @mardzix.bsky.social ! Could you share a high resolution image of the H&E and DAPI of the mouse embryo from your study? We would love to test our method on your dataset and that would help a lot!

I might be misunderstanding but maybe %load_ext autoreload helps?

IIRC dvc keeps a copy of your datasets every time you make changes, has storage space been an issue while changing big datasets a lot?

How much is known about MT reads in single nuclei data? Can one assume that the captured MT reads belong to mitochondria from the same cells as the nuclei?

Huh your link is also not working. I guess GEO is being weird

I watched ~5 YouTube videos on mixed effect models. Most of them said something like "including random slope and random intercept can make finding a fit difficult, random-intercept-only is the way to go."

Seemed fishy at the time, nice to find some more reading material here

@brukercorporation.bsky.social

Any idea?
Would love to use / cite this for a publication!