ZeroDivisionError: float division by zero when running ab initio

Troubleshooting
1

Hi,

I have been recently getting the following error when running ab initio model generation of a dataset with 66000 particles asking for 10 classes:

Traceback (most recent call last):
File “/datahdd/cryosparc/cryosparc-compute/sparcjob.py”, line 318, in
cuda_devs = current_experiment_params[‘cuda_devs’])
File “/datahdd/cryosparc/cryosparc-compute/abinit.py”, line 401, in abinit
**search_params)
File “/datahdd/cryosparc/cryosparc-compute/search.py”, line 359, in heuristicsearches
doscale = doscale, ignore_dc = ignore_dc, cuda_workers = cuda_workers)
File “/datahdd/cryosparc/cryosparc-compute/search.py”, line 245, in compute_Us
results.append(work(arg))
File “/datahdd/cryosparc/cryosparc-compute/search.py”, line 214, in work
IW.load(D, C, 1.0/N_R, 1.0/N_S, 1.0,
ZeroDivisionError: float division by zero

The error always happens at around iteration 550, although the exact iteration at which the error takes place varies between runs.

I have tried also asking for 9 classes, but got the same error (also around iteration 550). It seems to be somehow similar to a previous issue (ZeroDivisionError: float division by zero)

Has anyone managed to get around this error without having to lower too much the number of classes?

Thanks in advance!

2

Hi Rafael,

Thanks for reporting this. The issue is a bug in cryoSPARC that we’re trying to diagnose. It doesn’t always happen (I’ve run 10+ class jobs successfuly before) but when it does, there is not yet a workaround. It’s high priority, as soon as we fix it, it’ll be deployed.

Ali

3

I’ve seen this too - in my case, it only happened late in the piece, once several classes had become entirely empty

Oliver