Home

IceGraph is an end-to-end framework for building and deploying Graph Neural Networks for reconstruction and classification in IceCube. It covers the full path from raw detector files to a trained, deployable model through three workflows:

  • Processing converts events from source files to an ML-ready graph dataset.

  • Training fits a GNN to that dataset using the PyTorch framework.

  • Inference applies a trained model to new data to produce predictions.

All three are driven by YAML configuration and share a common substrate, so most of this documentation is organized around the framework’s objects and what each is responsible for.

To get started with scripting, see Usage.

Authorship and Disclaimer

IceGraph is authored by me, Taylor St Jean. AI was used (and will continue to be used) as a debugging and optimization tool during development, but the codebase, architecture, and all implementation decisions are my own work.