Skip to content
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

Feature request: Support vectorisation for Screen.reading histogram #351

Open
amylizzle opened this issue Feb 24, 2025 · 3 comments
Open
Labels
enhancement New feature or request

Comments

@amylizzle
Copy link

Screens are great outputs for training against, and doing them in batches would be preferred!

@jank324
Copy link
Member

jank324 commented Feb 24, 2025

Cheetah is fully vectorised. Unless there is a bug, it should therefore be no problem to get a batch of screen readings by, for example, supplying a vector of magnet settings with something like this:

segment = cheetah.Segment(
    [
        cheetah.HorizontalCorrector(angle=torch.tensor([1e-5, 2e-5, 3e-5]), length=torch.tensor(0.15)),
        cheetah.Drift(length=torch.tensor(1.0)),
        cheetah.Screen(resolution=(200, 100), is_active=True, name="my_screen"),
    ],
)
outgoing_beam = segment.track(incoming_beam)

segment.my_screen.reading.shape   # Should be (3, 100, 200)

@amylizzle
Copy link
Author

Odd, I went to do a batch of screens for a quad with a k1 set to a tensor(5,1) and got:

File [/opt/conda/lib/python3.12/site-packages/cheetah/accelerator/screen.py:252](https://apml1.dl.ac.uk/opt/conda/lib/python3.12/site-packages/cheetah/accelerator/screen.py#line=251), in Screen.reading(self)
    245 if self.method == "histogram":
    246     # Catch vectorisation, which is currently not supported by "histogram"
    247     if (
    248         len(read_beam.particles.shape) > 2
    249         or len(read_beam.particle_charges.shape) > 1
    250         or len(read_beam.energy.shape) > 0
    251     ):
--> 252         raise NotImplementedError(
    253             "The `'histogram'` method of `Screen` does not support "
    254             "vectorization. Use `'kde'` instead. If this is a feature you "
    255             "would like to see, please open an issue on GitHub."
    256         )
    258     image, _ = torch.histogramdd(
    259         torch.stack((read_beam.x, read_beam.y)).T,
    260         bins=self.pixel_bin_edges,
    261         weight=read_beam.particle_charges
    262         * read_beam.survival_probabilities,
    263     )
    264     image = torch.flipud(image.T)

NotImplementedError: The `'histogram'` method of `Screen` does not support vectorization. Use `'kde'` instead. If this is a feature you would like to see, please open an issue on GitHub.

@jank324
Copy link
Member

jank324 commented Feb 24, 2025

Oh, you're right. I forgot that when method="histogram" vectorisation is not supported. This is unfortunately a result of the histogram method in PyTorch not supporting vectorisation. If you set method="kde", vectorisation should work though.

Of course, as soon as PyTorch supports this, we would like to add it to Cheetah as well (or if someone has a good hand-made implementation).

@jank324 jank324 added the enhancement New feature or request label Feb 24, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
enhancement New feature or request
Projects
None yet
Development

No branches or pull requests

2 participants