NetBurst-essential

Standalone repository for running the NetBurst pipeline end-to-end:

• preprocess
• train
• analysis
• retrieval

What each module does

• src/preprocess/ - sparse time-series -> IBG/BI -> train/test/context splits
• src/train/ - model training, finetuning, inference, representation extraction
• src/analysis/ - clustering/CKA/anisotropy workflows and JSON-driven analysis pipelines
• src/retrieval/netburst_faiss/ - FAISS index build and retrieval evaluation

How to run

Follow module runbooks in this order:

1. src/preprocess/README.md
2. src/train/README.md
3. src/analysis/README.md
4. src/retrieval/netburst_faiss/README.md

What to expect

• Preprocess outputs under configured output paths (commonly <repo-root>/outputs/)
• Train checkpoints and evaluation artifacts
• Analysis CSV outputs and optional plot artifacts
• Retrieval FAISS index artifacts and query/distance CSVs

Citation

Paper link:

• https://arxiv.org/abs/2510.22397

BibTeX:

@misc{guthula2025netbursteventcentricforecastingbursty,
      title={NetBurst: Event-Centric Forecasting of Bursty, Intermittent Time Series}, 
      author={Satyandra Guthula and Jaber Daneshamooz and Charles Fleming and Ashish Kundu and Walter Willinger and Arpit Gupta},
      year={2025},
      eprint={2510.22397},
      archivePrefix={arXiv},
      primaryClass={cs.NI},
      url={https://arxiv.org/abs/2510.22397}, 
}