Implement Shower Maximum and/or First Interaction Depth reconstruction

[mescobargodoy]

Please describe the use case that requires this feature.
I am interested in reconstructing the Shower Maximum and/or First Interaction Depth of gamma-rays using the prototype Schwarzchild-Couder Telescope and VERITAS telescopes.
Machine learning is an avenue that can provide a better reconstruction than analytical methods. This would be part of my PhD project which specifically aims to measure the gamma-ray pair production cross section through the measurement of Shower Maximum and/or First Interaction Depth. I would like to implement this in the CTLearn framework.

Describe the solution you'd like
Add an extra task to perform showermax and/or first_int_depth.

Additional context
The idea I have in mind is sort of replicate what is already done with the existing reconstruction tasks that perform regression.
ctlearn is relatively new to me so f you could point me to a good starting point I would greatly appreciate it. Let me know if there's anything else I should be aware of.
Thanks!

[TjarkMiener]

Thanks @mescobargodoy I recently added the impact reconstruction (see #257). The other reconstructions should be added on top of this branch in a similar way.

[mescobargodoy]

Thanks @TjarkMiener! I have a working version here https://github.com/mescobargodoy/ctlearn/tree/xmax
I've done some preliminary tests and the tools ctlearn-train-model & ctlearn-predict-mono-model seem to be working fine.
I just had a few questions.

• Would the stereo combiner require a ctapipe.container for xmax?
• I can't seem to find the predicted xmax in the output file. I am using the read_table method from ctapipe.io to inspect the file. I was wondering if this is the correct way to open the file and view it's contents or is there a different method to use here? I checked the groups as well and I was expecting to see a Table under /dl1/event/telescope/xmax or something similar but I don't see one either.

Thanks again!

[TjarkMiener]

Hi @mescobargodoy, I had a quick check over the code. It is looking good. What command are you using for the prediction tool? Does it includes --dl2-telescope? I think the default is False, so it simply does not store the prediction in the output file for the telelscope-level. I think you would also need to disable the DL2 subarray prediction via --no-dl2-subarray since it is not implemented.

Would the stereo combiner require a ctapipe.container for xmax?

It would need to be added to the enum.Flag of the ReconstructionProperty. I do not know if ctapipe is planning to expand the enum in the future.

[mescobargodoy]

Hi @TjarkMiener,
Thanks for pointing this out and for the additional information! I believe it was an error on my end as I was probably looking at the incorrect file so apologies. It seems by default --PredictCTLearnModel.dl2_telescope=True and after doing a few more checks indeed the reconstructed Xmax is written to the table as CTLearn_tel_xmax.

Everything seems to be working fine on the xmax branch. Thanks for the help! Seeing that the impact branch has been requested to merge I am assuming there will be more changes that will need to be added in the future so I don't plan to make a pull request at the moment. If you believe there's anything else that should be included let me know and I can take a look at it.