Ab initio errors

Hi, I have a dataset where ab initio almost never works (19 out of 20 attempts). Nearly all jobs terminate with either Assertion error or Value error (NaN values detected).

The dataset passes Check particles without detecting anything unusual. Same particles work just fine in NuRefine, 2D classification and Heterogenous refinement. So there is nothing wrong with dataset itself.

The error in ab initio comes after the first 49 or 99 iterations. Reducing the dataset from 500k to 50k particles did not help. Box size is 120 pixels at the moment, but it was failing with smaller box size too.

I tried two versions of cryosparc: 4.6.2 (on a cluster) and 4.5.3+240807 (standalone).

I will be happy to share the data, if that could help to troubleshoot.

Peter

Thanks @peter.cherepanov for reporting. For an ab initio job that failed with a NaN-related error on CryoSPARC v4.6.2, please can you

  • confirm that the error also occurs with the Cache particle images on SSD option disabled
  • email us the job report for a failed ab initio job where the Cache particle images on SSD option had been disabled
  • email us the job report of the corresponding Check Particles job

Thank you for your response, wtempel,
Yes, the job fails also when SSD is disabled. Here is the failed job report:

it failed after 50 iterations, with “ValueError: Detected NaN values in engine.compute_error.”

Also, here is a report from checking particles in the whole dataset (this is from a different computer, since for some reason I cannot run Check particles job on the cluster version at the moment):

Peter

Thanks @peter.cherepanov. I fetched the reports, in case you wish to “unshare” the reports.

I also see the same issue - ab initio reproducibly failing with NaN errors on a certain (confirmed good) particle set and resolution range. Ab initio jobs on the same particle set with altered resolution range finish without issue.

I will try that thank you.