Open nicolasaunai opened 5 months ago
this might be linked to #824. We may not be able to reduce the number of calls to regrid but we may not have to explicitly regrid (i.e. change the patch distrib) as often if there was not as much noise in the tags. Perhaps the high frequency evolution of the patch distribution comes from noise in tags. Tags would, on average, be at the same location, but slightly different, which results in computing slightly different boxes and then effectively regriding.
At the moment the refinement is triggered a bit easily and creates spurious patches here and there especially from L0 to L1.
Example of L1 patches :
This issue is about adjusting a bit the threshold so that it does less that. Let the tagging the threshold accessible from python (see Nathan's work) so that study can be done easily without recompiling (and useful for end user as well). Assess the effect of increasing the threshold in various simulations of Harris type. Assess the effect of less noisy patch creation on the performance and scalability.