Adding EQDSK support #391
Adding EQDSK support #391
Conversation
… and build_sim files. The flux surface averaging is done using contourpy and RectBivariate spline interpolation of the PSI 2D grid. Note, there appear to be systematic differences in the shaping parameters (R inboard, R outboard, triangularity).
|
Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA). View this failed invocation of the CLA check for more information. For the most up to date status, view the checks section at the bottom of the pull request. |
torax/geometry_loader.py
Outdated
| @@ -58,6 +59,11 @@ def _load_fbt_data(file_path: str) -> dict[str, np.ndarray]: | |||
| """Loads the data from a FBT-LY file into a dictionary.""" | |||
| return scipy.io.loadmat(file_path, squeeze_me=True) | |||
|
|
|||
| from eqdsk import EQDSKInterface | |||
| def _load_eqdsk_data(file_path: str) -> dict[str, np.ndarray]: | |||
| eq = EQDSKInterface.from_file(file_path, no_cocos=True) # should probably handle COCOS shenanigans... | |||
There was a problem hiding this comment.
Is this something we need to handle here? Or maybe turn it into a formal todo
There was a problem hiding this comment.
This needs a formal TODO, not sure how the torax team wants to handle COCOS in the future.
torax/geometry_loader.py
Outdated
| from eqdsk import EQDSKInterface | ||
| def _load_eqdsk_data(file_path: str) -> dict[str, np.ndarray]: | ||
| eq = EQDSKInterface.from_file(file_path, no_cocos=True) # should probably handle COCOS shenanigans... | ||
| return eq.__dict__ |
There was a problem hiding this comment.
Use dict(eq) but also maybe rename to something like loaded_data or similar as eq is generally used for equals
There was a problem hiding this comment.
dict(eq) did not work, therefore I have kept it as eq.__dict, except renaming as suggested.
torax/geometry.py
Outdated
| # Gathering area for profiles | ||
| areas, volumes = [], [] | ||
| R_inboard, R_outboard = [], [] | ||
| flux_surf_avg_1_over_R2_eqdsk = [] # <1/R**2> |
There was a problem hiding this comment.
nit: the underscore eqdsk is probably not necessary here?
To save vertical space etc is it worth preallocating the numpy arrays and then assigning in rather than needing to create the arrays and then turn them into arrays?
There was a problem hiding this comment.
I think the underscore is useful as the from_chease method has similar syntax for the FSA quantities.
|
Can you also sign the CLA please. We will need to do a bit of work internally to bring this in and the dependencies internally but will work on that in the meantime |
… arrays in the contour integration
|
Looks good, are you able to resolve the conflicts so that we can try it out |
|
|
||
| # ---- Interpolations | ||
| q_interp = scipy.interpolate.interp1d(psi_eqdsk_1dgrid, eqfile['qpsi'], kind='cubic') | ||
| psi_spline_fit = scipy.interpolate.RectBivariateSpline(X_1D, Z_1D, psi_eqdsk_2dgrid, kx=3, ky=3, s=0) |
There was a problem hiding this comment.
out of curiosity, did you check robustness of the results to the hyperparameters of this method?
|
|
||
| # set epsilon based on floating point precision of psi grid | ||
| # NOTE: this is hacky and should be replaced with a more robust method | ||
| epsilon = 1e-7 |
There was a problem hiding this comment.
nit: can use eps from the constants.Constants dataclass
There was a problem hiding this comment.
Was this generated directly from the existing ITER_hybrid CHEASE file?
If so, and if it doesn't break the EQDSK file, would be good to mention this in a header
There was a problem hiding this comment.
if all else apart from geometry is identical to test_iterhybrid_predictorcorrector, then can shorten this file as well as making the provenance clearer, by copying the CONFIG from iterhybrid and then modifying it. See https://github.com/google-deepmind/torax/blob/main/torax/tests/test_data/test_iterhybrid_rampup_restart.py for an example
|
Thanks, looks really good. Made some minor comments after another run through. After these, and merging with main, we'll bring it in internally and push it from there, next week. Thanks again for this! |
A draft for EQDSK support by modifying the geometry, geometry_loader, and build_sim files.
A corresponding geometry file is contributed to
data/third_party/geo(credit to Antti Salmi for the help in geometry file generations).Additionally, an example configuration is made based off (copied from) the iter hybrid scenario and basic example already provided. But the PR can likely do without the
eqdsk_exampleconfig... the example takes a long time to run (at least it runs :D)...The flux surface averaging is done using contourpy and RectBivariate spline interpolation of the PSI 2D grid. Note, there appear to be systematic differences in the shaping parameters (R inboard, R outboard, triangularity).
Somewhere the requirements have to be updated to include the TSVV-15's eqdsk and contourpy
See #373 for more details and discussion.