Closed fresleven closed 3 months ago
I also changed the SWE Nonlinear loss function that I previously put since I noticed that there was a slight problem with the indexing last time (it should've gone from [1:-1] for all values since we ignore the boundaries, but there were some values where it went from [0:-2] (i.e. for 2 * nt - 2
. Thus, I made it simpler by just indexing at the end.
Hi, I just wanted to know when you would be able to review my code?
Hey @fresleven, I have been looking at the code and it looks good but would like to try it out. Do you have a dataset already made you could send? Even if its a smaller subset. I was running into issues installing Dedalus.
Hi, I just made a google drive with 100 samples:
https://drive.google.com/drive/folders/10xeaTHRMLKnkPW5uXXEbBDktPgdQ5qW2?usp=sharing
and you need to put it in the correct mhd_pino/mhd_data/simulation_outputs_Re250
directory as well
you may need to change the config files based on how much training/testing files there are, like in lines ~50 you may need to change the num
, num_test
, and num_train
ones
Also, please let me know what issues you were having with Dedalus and I can try and see to replicate/fix them soon. I was using conda, and that worked for me.
also, Professor Huerta notified me that the deadline is by 5pm EST today. I know it's quite short, but hopefully could you complete the review by then?
/blossom-ci
Modulus Pull Request
Description
Added a physics-informed MHD example and altered the loss function of SWE Nonlinear PINO to have the correct indexing.
Checklist
Dependencies
Uses tensorly, tltorch, plotly, and dedalus. More information on how to install these packages is included in the README in examples/cfd/mhd_pino/