The code is for our paper "FluxEV: A Fast and Effective Unsupervised Framework for Time-Series Anomaly Detection" and this paper has been accepted by WSDM 2021. [中文解读]
😉 About Name: "Flux" means the Fluctuation, "EV" denotes the Extreme Value.
💫 Good News: FluxEV has been integrated into OATS, a convenient Python library providing various approaches to time series anomaly detection. Thanks for their awesome work!
- numpy
- numba
- scipy
- pandas
- scikit-learn
- matplotlib (plot)
- more-itertools (plot)
- KPI.
- Competition Website Link: http://iops.ai/dataset_detail/?id=10 (
⚠️ Seems like the official competition website is down!) - Tsinghua Netman GitHub Link: https://github.com/NetManAIOps/KPI-Anomaly-Detection/tree/master/Finals_dataset
- Yahoo.
- Official Website Link: https://webscope.sandbox.yahoo.com/catalog.php?datatype=s&did=70
preprocessing.py
:
- Fill the missing points for KPI dataset.
spot_pipe.py
:
- SPOT function is modified to be a part of FluxEV for streaming detection;
- MOM(Method of Moments) is added as one of parameter estimation methods;
- For the original code, please refer to SPOT (Streaming Peaks-Over-Threshold)
eval_methods.py
:
- The adjustment strategy is consistent with AIOps Challenge, KPI Anomaly Detection Competition.
- The original evaluation script is available at iops.
main.py
:
- Implement streaming detection of FluxEV.
python main.py --dataset=KPI
python main.py --dataset=Yahoo
@inproceedings{li2021fluxev,
title={FluxEV: A Fast and Effective Unsupervised Framework for Time-Series Anomaly Detection},
author={Li, Jia and Di, Shimin and Shen, Yanyan and Chen, Lei},
booktitle={Proceedings of the 14th ACM International Conference on Web Search and Data Mining},
pages={824--832},
year={2021}
}