I am running a local refinement in cryoSPARC 4.2.1, with ‘Optimize per-particle scale’ activated. It seems to help a lot, the resulting density is at higher resolution and looks reasonable.
Estimating particle scales gives this output:
The problem is that after a few iterations, I get this message in the log:
Rescaling optimal scales and reconstruction result to preserve median 1.0. Rescaling factor is nan
After this, the density disappears and the reconstruction just produces empty boxes:
Refining scales does seem unstable in local refinement.
I don’t have a solution to this really, but I wonder if you kill your job after it has done a couple of iterations (before it becomes unstable) and mark it as complete, if you could then use the output particles for a round of local refinement with fixed scales (using the scales you have already refined)?
Hi @shuber,
We aren’t sure what might be causing this, but to help our investigation, could you send us (either on the forum, or via DM) the Job error report?
Michael
We can confirm that this error sometimes occasionally happens also in version 4.6.2. Please include checks in the code for NaN values after the step of rescaling the scale factors to preserve the median.
Daniel is referring to one of my local refinement jobs. I will rerun it with scale factor refinement on and send the job error report when it crashes.
Some more observations: in my case, the local refinement job crashed consistently at the end of iteration 0 (so after the first full pass through the particle images), exactly because the reconstruction at this point is an empty box.
A “check particles” job validated that the subtracted particle images used for this job are good, and a reconstruction-only job with the same particle stack copied to cache completed fine, suggesting that the copy to the cache is faithful to the original data.
@mmclean I just sent you the report of such a crashed job in your direct messages. I hope this helps!
This seems to be the same problem as in this other discussion: Local refinement error (cryoSPARC v4.5.3)
Thank you Guillaume. Job report received, we will take a look.
–Michael
