[aspect-devel] Postprocessing scripts
jperryh2 at uoregon.edu
Tue Oct 30 11:32:44 PDT 2018
Perhaps a repository of paraview filters would be useful. It's possible
to save a "pipeline" as a filter, and expose certain variables to the
user through it.
Marine: I agree with Wolfgang that this is not an efficient way to do
things, but it does come in handy sometimes. For what it's worth, here's
a script to convert vtk output to csv, which you can then handle with
the usual python tools:
from paraview import simple as pvsimple
for ifname in glob('output/solution/*.pvtu'):
ofname = ifname.rsplit('.', 1)+'.csv'
reader = pvsimple.OpenDataFile(ifname)
writer = pvsimple.CreateWriter(ofname, reader)
writer.FieldAssociation = 'Points'
I also highly recommend compressing those files, because they'll
probably be enormous:
import gzip, os, shutil
with open(ofname, 'rb') as f_in, \
gzip.open(ofname+'.gz', 'wb') as f_out:
numpy (and probably Pandas and whatnot) automatically recognize gzipped
data and treat it transparently as a plain csv file.
Hope that helps someone,
On 10/30/18 10:11 AM, Wolfgang Bangerth wrote:
> I'm not sure anyone ever answered this email -- so let me do this 3
> months late...
>> But to be able to do a good comparisons with existing results, I'll need
>> things like horizontal averages (of composition, temperature,
>> velocities) and difference between the field and the horizontal average
>> (of composition, temperature). But the horizontal here is average over a
>> sphere, so I am guessing the post-processors are not so easy. (to get
>> average radial profiles)
> This is actually what our 'depth average' postprocessor does -- average
> over all points at a certain depth where depth is defined differently
> for each geometry.
>> I'm definitely not used to paraview, and I'm more used to do all my data
>> processing by hand... So a good tutorial on how to implement that with
>> paraview or equivalent, with a scripting (I usually have several dozens
>> of runs to process at a time) would be nice :-) But so far, I think that
>> a way to obtain a list of the points and values in the volume (without
>> duplicates), and maybe something to do the interpolation at required
>> points would be nice.
> This isn't going to be efficient. There may be tens of millions of
> points in a simulation, and they are stored by maybe hundreds of
> processors, and the interpolation between individual points is really
> quite complicated if you have a non-trivial geometry. There are really
> only two efficient ways to deal with this much data: by outputting the
> data in specialized file formats for visualization, and by implementing
> postprocessors within ASPECT. I know that you'd like to do this by hand
> in some scripting language, but don't do it -- you won't be happy this
> way if the data just becomes too large.
More information about the Aspect-devel