StreamFP

StreamFP is the implementation repository for fingerprint-guided data selection in efficient stream learning. The canonical maintained repository is DataSysResearch/StreamFP. It was migrated from the historical intellistream/StreamLearning repository; that old location is retained only as provenance and may redirect here.

StreamFP belongs to DataSys because its primary concern is online data selection, replay-buffer maintenance, and update efficiency over evolving data, not application-level agent or workflow orchestration.

Project status

This repository is a research artifact for reproducing the StreamFP experiments. It is not presented as a general-purpose training framework.

Publication

This repository accompanies the following paper:

• Changwu Li et al. "StreamFP: Fingerprint-guided Data Selection for Efficient Stream Learning." The ACM Web Conference 2026 (WWW 2026). DOI

Setup

• Install miniconda
• conda env create -f environment.yml
• conda activate sl
• Install fastmoe library: https://github.com/laekov/fastmoe/blob/master/doc/installation-guide.md

Datasets

• Create a folder data/
• Clear10, Clear100: retrieve from: https://clear-benchmark.github.io/
• Stream51: retrieve from: https://github.com/tyler-hayes/Stream-51
• CORe50: sh core50.sh

Training

All commands should be run under the project root directory. The scripts are set up for 1 GPUs but can be modified for your hardware.

sh experiments/clear10.sh
sh experiments/clear100.sh
sh experiments/core50.sh
sh experiments/stream51.sh

License

No repository-wide license has been declared. Contact the maintainers before redistributing or reusing the code, and review the terms of bundled or imported third-party components separately.