Skip to content

[PyWrapper] Added hybrid PyTorch-SU2 coupling example - #71

Open
ayush4874 wants to merge 3 commits into
su2code:masterfrom
ayush4874:feature/add-pytorch-example
Open

[PyWrapper] Added hybrid PyTorch-SU2 coupling example#71
ayush4874 wants to merge 3 commits into
su2code:masterfrom
ayush4874:feature/add-pytorch-example

Conversation

@ayush4874

@ayush4874 ayush4874 commented Dec 10, 2025

Copy link
Copy Markdown

This PR adds a standalone example demonstrating how to couple SU2 with PyTorch using the Python Wrapper.

Files:

  • train_online.py: Driver script for bidirectional coupling.
  • inv_NACA0012.cfg: Configuration file.
  • mesh_NACA0012_inv.su2: Mesh file.

Context:
Migrated from su2code/SU2#2638 and the tracking issue su2code/SU2#2637 as per discussion with @bigfooted.
This example serves as a reference for users implementing online surrogate modeling or PIML workflows.

@ayush4874

ayush4874 commented Dec 19, 2025

Copy link
Copy Markdown
Author

Hi @bigfooted,

I've successfully migrated the tutorial files here and opened the corresponding documentation PR in the website repo (su2code/su2code.github.io#189).

Regarding the regression test: I am ready to implement this, but I have a technical concern. This tutorial depends on torch, and I noticed that the current regression.yml in the main repo doesn't seem to include PyTorch in the environment.

Should I:

  1. Structure the test to be skipped if torch is not found?
  2. Or would you prefer to merge this as a "manual-only" tutorial for now to avoid modifying the CI environment?

I’m available to implement whichever path you prefer before the holiday break. Thanks!

@bigfooted

Copy link
Copy Markdown
Contributor

Let's not add pytorch to the regression test, it is very large and will increase the size of the docker container by a lot.

@ayush4874

Copy link
Copy Markdown
Author

Thanks for the feedback @bigfooted. That makes sense, adding the full PyTorch dependency to the regression test environment significantly increases the container size for a single tutorial.

Plan:

  1. I will revert the changes to the environment/Docker files (removing PyTorch).
  2. I will keep the tutorial files (scripts/configs) so users can run it if they have PyTorch installed locally.
  3. I will add a note to the tutorial documentation specifying the manual dependency.
  4. I will remove this specific case from the automated regression suite to prevent CI failures.

I'll push these updates shortly.

@ayush4874

Copy link
Copy Markdown
Author

@bigfooted it's ready for review, can you have a look?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

2 participants