-
Notifications
You must be signed in to change notification settings - Fork 27
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Fix some plot issues when NFP differs from one for objects, or when passed-in phi exceeds 2pi/nfp #1204
Conversation
…assed-in phi exceeds 2pi/nfp
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #1204 +/- ##
=======================================
Coverage 95.33% 95.33%
=======================================
Files 90 90
Lines 22699 22700 +1
=======================================
+ Hits 21640 21642 +2
+ Misses 1059 1058 -1
|
| benchmark_name | dt(%) | dt(s) | t_new(s) | t_old(s) |
| -------------------------------------- | ---------------------- | ---------------------- | ---------------------- | ---------------------- |
test_build_transform_fft_lowres | +2.14 +/- 6.84 | +1.10e-02 +/- 3.51e-02 | 5.24e-01 +/- 3.4e-02 | 5.13e-01 +/- 8.8e-03 |
test_build_transform_fft_midres | +0.55 +/- 3.63 | +3.27e-03 +/- 2.17e-02 | 6.01e-01 +/- 1.0e-02 | 5.98e-01 +/- 1.9e-02 |
test_build_transform_fft_highres | -0.55 +/- 3.81 | -5.48e-03 +/- 3.81e-02 | 9.95e-01 +/- 2.4e-02 | 1.00e+00 +/- 2.9e-02 |
test_equilibrium_init_lowres | +5.83 +/- 4.14 | +2.18e-01 +/- 1.54e-01 | 3.95e+00 +/- 1.5e-01 | 3.73e+00 +/- 3.5e-02 |
test_equilibrium_init_medres | +0.15 +/- 3.59 | +6.12e-03 +/- 1.47e-01 | 4.11e+00 +/- 8.7e-02 | 4.10e+00 +/- 1.2e-01 |
test_equilibrium_init_highres | +0.27 +/- 2.43 | +1.47e-02 +/- 1.32e-01 | 5.44e+00 +/- 9.6e-02 | 5.43e+00 +/- 9.1e-02 |
test_objective_compile_dshape_current | +0.97 +/- 2.84 | +3.69e-02 +/- 1.08e-01 | 3.85e+00 +/- 1.0e-01 | 3.81e+00 +/- 2.8e-02 |
test_objective_compile_atf | +0.21 +/- 1.88 | +1.66e-02 +/- 1.46e-01 | 7.81e+00 +/- 7.9e-02 | 7.79e+00 +/- 1.2e-01 |
test_objective_compute_dshape_current | -1.36 +/- 1.71 | -4.72e-05 +/- 5.94e-05 | 3.42e-03 +/- 3.1e-05 | 3.47e-03 +/- 5.1e-05 |
test_objective_compute_atf | -0.16 +/- 2.41 | -1.61e-05 +/- 2.47e-04 | 1.02e-02 +/- 2.1e-04 | 1.02e-02 +/- 1.3e-04 |
test_objective_jac_dshape_current | +3.88 +/- 9.15 | +1.50e-03 +/- 3.53e-03 | 4.01e-02 +/- 1.7e-03 | 3.86e-02 +/- 3.1e-03 |
test_objective_jac_atf | -0.83 +/- 3.58 | -1.59e-02 +/- 6.82e-02 | 1.89e+00 +/- 3.4e-02 | 1.91e+00 +/- 5.9e-02 |
test_perturb_1 | +0.94 +/- 1.05 | +1.14e-01 +/- 1.27e-01 | 1.22e+01 +/- 8.0e-02 | 1.21e+01 +/- 9.9e-02 |
test_perturb_2 | +0.80 +/- 1.45 | +1.37e-01 +/- 2.48e-01 | 1.72e+01 +/- 2.3e-01 | 1.71e+01 +/- 8.2e-02 |
test_proximal_jac_atf | +0.25 +/- 1.15 | +2.00e-02 +/- 9.29e-02 | 8.07e+00 +/- 6.6e-02 | 8.05e+00 +/- 6.5e-02 |
test_proximal_freeb_compute | -0.68 +/- 0.87 | -1.25e-03 +/- 1.61e-03 | 1.83e-01 +/- 8.1e-04 | 1.84e-01 +/- 1.4e-03 |
test_proximal_freeb_jac | -0.73 +/- 2.73 | -5.38e-02 +/- 2.02e-01 | 7.35e+00 +/- 6.9e-02 | 7.41e+00 +/- 1.9e-01 |
test_solve_fixed_iter | +0.82 +/- 61.16 | +3.98e-02 +/- 2.97e+00 | 4.89e+00 +/- 2.1e+00 | 4.85e+00 +/- 2.1e+00 | |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For your first point, isn't the root issue #1180?
Coordinates are not periodic functions, so we shouldn't treat them as such in the compute functions or constructing grids.
Here is my proposed solution:
- Change
LinearGrid
to not mod$\zeta$ by$2 \pi / \text{NFP}$ and also not mod$\theta$ by$2 \pi$ . - In
LinearGrid
, if the coordinates span multiple periods, that is ifnp.ptp
($\zeta$ )$> 2 \pi / \text{NFP}$ ornp.ptp
($\theta$ )$> 2 \pi$ , then set the attributes_spacing
and_weights
toNone
.
Reasoning:
- Periodic functions should not be affected by this logic. This would be a bug fix for non-periodic functions.
- Those attributes are quadrature weights for an integration over a geometry of NFP periods. When coordinates span a different length than that, one is effectively defining multiple copies of a volume or surface. We always want to compute an integral over NFP periods, so we modify the weights accordingly.
In theory one could always modify the quadrature weights so that the integral is over the domain we want regardless of the coordinate span, but the API ofLinearGrid
is too broad to make that simple.
Throw an error when |
phi = parse_argname_change(phi, kwargs, "zeta", "phi") | ||
errorif( | ||
not np.allclose([thing.NFP for thing in eqs], eqs[0].NFP) and phi is None, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@dpanici Maybe we should change this to
not np.allclose([thing.NFP for thing in eqs], eqs[0].NFP) and (phi is None or isinstance(phi, numbers.Integral)),
Because integer phi
is still ambigious.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
ah, what does it do rn if phi integral is passed in? btw 0->2pi or 0->NFP of the first thing?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It uses 0 to 2pi/nfp of the last thing.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also, doesn't throw any error or warning.
plot_surfaces
(andplot_section
andplot_boundary
) whenphi
is supplied which exceedsNFP
, the grid used for the map coordinates (a lineargrid withNFP=eq.NFP
would truncate thephi
. This is fine unless the new truncated phi has some duplicates. The grid is still a meshgrid of the supplied rho, theta, zeta, however, then thegrid.num_zeta
value, which gives the unique zeta values, might not correspond to the original length ofphi
. This causes reshaping errors like those seen in TypeError in 'plot_comparison' #1202. These grids though dont need to have the NFP arg at all as they always supplyphi
as an array (instead of supplyingN
), so we can set them to1
to avoid this issieplot_comparison
andplot_boundaries
, there is some ambiguity on what should happen by default when things of differing NFP are passed in. This changes to by default, throw an error if multiple nonaxisymmetric objects with difering NFP are passed in (as it is not clear what phis to do, and even then the plot title "phi*NFP/2pi" is ambiguous (which NFP does it refer to?)). If there are differing NFPs but the only ones with nonzero NFPs have the same NFP, and the remaining are axisymmetric objects, it will change the axisymmetric object NFP to match the nonaxisymetric ones (we can do that and it does not affect the plot since they are the same at every zeta).The change in
plot_boundaries
behavior is something I am not 100% sure on, what do people think it should be? we could leave as is and just plot a bunch of XS, maybe with the phi being determined inside of the for loop instead of outside, so that if the default is used you get (0,2np.pi,4,endpoint=False). The only weird thing is now you have a bunch of XS where they are at differing angles and it is not clear how they should be compared to eachother.Resolves #1202