Machine learning pipeline for reconstruction in neutrino telescopes. Contains code for running SSCNN and neutrino telescope super-resolution on Prometheus events, as described in arXiv:2303.08812 and arXiv:2408.08474. The working environment used to implement, run and test this code are as follows:
- Python 3.11.2
- PyTorch 2.2
- PyTorch Lightning 2.1
- CUDA 12.1
- MinkowskiEngine (for SSCNN) 0.5.4
- Segmentation Models PyTorch
- Weights&Biases
- NumPy
- Awkward Arrays
You can define a configuration file and run it with python train.py -c config.cfg.