[Paraview] How to Average Transient Data

[John Biddiscombe]

Kent

I have got a vtkTemporalInterpolationFilter which is the first class in 
what I hope will become a new pipeline model. It maintains a cache of 
datasets each of which corresponds to a timestep of the data. It could 
be simply modified to make it a vtkTemporalAveragingFilter which would 
do what you want.

I've made some changes to the executives to support the temporal 
requests - but am still only half finished and would like to spend 
another couple of weeks on it before I put it on our public svn server. 
Can you wait that long? (also If you have specific requests above your 
initial email, then please ask)

The basic idea is that any Reader (or pipeline section) that is Time 
enabled (currently my NetCDF, HDF5-Particle), plus Ensight, Exodus, vtk 
xml I think(others?) will provide a requested timestep from the 
available timesteps on disk/file. The TemporalPipeline manages the 
RequestInformation and RequestData, RequestXXX phases by looping the 
upstream pipeline and placing the data in a cache. The algorithm that 
subclasses the TemporalPipeline can request TimeStep N (or more 
preferably for example TIME_VALUE T=4.6), the pipeline will then loop 
and fetch datasets from T=4, T=5 and produce an interpolated dataset 
coresponding to T=4.6. The algorithm can request a timestep and let the 
cache return it, generate it, or fetch it if not present. When a 
timestep is requested which does not use any cache'd datasets, the 
cache'd ones are dropped (could be modified to use a better algorithm 
here such as MRU N datasets).

In the case of the TemporalAveragingFilter, you would override the 
Algorithm to do an average instead of an interpolation, and also specify 
how many datasets you need for each timestep. Averaging by 2,3,4....N 
would require only one parameter change.

I will experiment by implementing the averaging algorithm.

One catch. The whole process is going to be unstable if the datasets 
change geometry between timesteps. Currently, I can average all the 
scalars and points, if they are invariant under the number present in 
each timestep. If they change betweeen steps, then we are in trouble.

I will keep you informed of progress and send code as soon as I'm 
confident that all is working stably.

JB