Hi cryoSPARC users Community,
I was wondering if there are any checkpoint to continue a job if for example this job fails …
(like in RELION)
is it possible to think about it for the next update if it is not possible please ?
Also finding a GUI option to continue a job from a checkpoint super useful. 3D classification without alignment jobs are long and when they fail (e.g. cluster instability) one would want to continue from where it was left of.
At least this has been requested before but I am not sure the devs have released this functionality yet in the current version. It’s not just the waiting, having to rerun long jobs is such a waste of compute resource.
Welcome to the forum @Nadia68 @dimkol94 . Please see Continue job option for refinements - #2 by wtempel under a related topic.
Hi @wtempel,
I understand the reasoning here - but I would just point out that checkpointing is not just helpful to recover from crashes/failures (though it definitely is that!).
It also adds a huge amount of flexibility to refinements and classifications that then does not need to be explicitly added. For example, one can start off local refinement with a wide search range, then tighten it up after continuing from a checkpoint. Or run 3D classification initially with a low target resolution, then increase that for subsequent iterations. Or perhaps start off without a highpass filter, then add a highpass filter once coarse alignments/class assignments are established. These approaches are very helpful in relion, and I miss them in cryosparc. Just my two cents
Cheers
Oli