[release/v2-stable] Troubleshooting

[jpark27]

Dear @EliHei2

Hi, Elias! My apologies for keep bugging you.
I happen to run all my samples under "release/v2-stable" but one dataset has following issue:

0) Datasets are 5K prime with boundary stain, so I used following command:

1) Pipeline run without error messages
2) Output (segger_segmentation.h5ad) only had ~500 cells in problematic case when another sample with same batch has ~1M cells

Do you happen to have similar experiences with some of your/other dataset before? Not sure which parameter I should touch in this case scenario. I shared my input/output of Xenium data (through your DKFZ e-mail address) and wonder if you can give some advices.

best wishes,
J

[EliHei2]

hey! thanks for reporting this, first if you could switch to v2-incremental it would be great, it's a debugged verison. but for this specific case, I would suggest to rerun again and it should work out, a tip is to run only the segger predict step, if your logs look nice (look up segger plot --quick). if the training curves look off you should run segger segment form scratch, but normally this would resolve, I'm inspecting currently what's the main cause of this.

[jpark27]

Thanks, @EliHei2! I will do.
Just for sanity check, v2-incremental installed as v0.1.0 when I git cloned. is it just typo? segger segment command structure also seem slightly different from v2-stable:

/project/IBD_CD4/jpark/segger_incremental$ cat requirements.txt 2>/dev/null || cat pyproject.toml | head -60
[project]
name = "segger"
version = "0.1.0"