AI energy-recon for SiPM-on-tile ZDC

Adds point-cloud AI-based energy reconstruction for the SiPM-on-tile version of the ZDC. This also changes the energy of the neutrons used in this detector's benchmark from 10 GeV to 20 GeV, since 10 GeV is outside of the range of validity for the model used for this reconstruction.

requested review from @whit

added 1 commit

• 7af04067 - Update run_zdc_sipmontile.sh

Compare with previous version

added 1 commit

• 145ede39 - Update run_zdc_sipmontile.sh

Compare with previous version

Dear @whit , In order for this benchmark to run, it is necessary to activate the conda environment specified in benchmarks/far_forward/tensorflow.yml in order to have all of the libraries used in the AI. It appears that conda is not installed in the environment that the benchmarks are running in. Let me know if it's possible to get conda installed in that environment, so that these libraries can be used.
Thanks, Sebouh

@sebouh137

What libraries do you need? Ideally, these would installed in eic_container which builds the container images used by eic-shell. You could also try installing what you need in the CI job if it is lightweight. I see you are using tensorflow? If you are not too committed to that library, I know there are lightweight python libraries that do the same things.

Conda is usually for setting up python environments. Not sure it makes sense to use it here. But could be useful for testing.

@whit I don't remember which of the libraries referenced in that yaml file are required for reconstruction and which ones are only used in training the model. I will ask my coworkers if the lightweight alternatives to tensorflow could be used for reconstruction (even if we'd continue to use tensorflow in training the models). After discussing these things with my coworkers, we will put in a pull request in eic_container in order to include the relevant set of libraries, so that we can proceed with this merge request. Thanks, Sebouh