Add scripts and files to process Vi's screen#14
Conversation
We need to lose the references to each future after we gather its result, or dask will keep it in-memory in a worker, past cancellation. See: dask/dask-jobqueue#122 (comment)
|
@GenevieveBuckley please annotate the bits of code that need clarification and I'll add comments, as I think these scripts should be used as templates to generalise to new screens, so they should be well documented. |
| # SBATCH --mail-user=juan.nunez-iglesias@monash.edu | ||
| # SBATCH --mail-type=END | ||
| # SBATCH --mail-type=FAIL | ||
| /home/jnun0003/.conda/envs/mic/bin/python ./single-process-vi.py |
There was a problem hiding this comment.
Same again, scripts in shared spaces would be better.
|
I forgot to include the output, which made me very happy (after seeing many outputs with full tracebacks and half-finished progress bars =P): Needless to say I don't know yet what those "Workers don't have promised key", but it doesn't seem to have affected any feature computation. |
|
I think now that UMAP is being included more routinely in these scripts (eg: c236e06) we should update |
|
Hmm, I thought I'd responded to this, sorry, my bad! Thoughts: when we pull these things into microscopium proper, we can update the requirements. As we've discussed recently, one thing I want to do pretty short term is to go from images + info-csv + config file to the full feature computation and dimensionality reduction. That PR is the place to update the requirements. |
|
btw assuming you have no further questions I think this is ready to merge. |
Uses dask-jobqueue for the main workflow.