Hi,
When performing heterogenous refinement with identical references, a randomization step is carried out in the initial iterations to break the symmetry of the identical references.
Presumably some kind of random seed is used to initialize this randomization process. Is it the same as the random seed provided in the job inputs?
The reason I ask is that when I start several cloned jobs with the same random seed (testing GPU performance and doing some troubleshooting), I get wildly different class percentages for jobs that have the same references, and the same random seed.
If I start with non-identical references this is not the case. This makes me wonder whether maybe a different random seed is being used during class randomization vs the one provided in the inputs?
Cheers
Oli