ADTK: An open-source Python toolkit for anomaly detection in...
summary
ADTK claims to be an open‑source Python toolkit that supplies ready‑to‑use building blocks — detectors, transformers, aggregators — with a unified API so practitioners assemble rules‑based time‑series models like Lego. The demo runs real pipelines (seasonal decomposition, rolling aggregates, autoregression, PCA wrappers), shows integration with scikit‑learn, provides evaluation for interval events (IoU/coverage thresholds) and time‑aware train/test splits, and is already applied to IoT monitoring workflows for streaming anomaly detection.
AI-written summary · about
chapters (19)
- 0:00ADTK introduction
- 1:07Anomaly detection in time series
- 3:12Time series anomaly challenges
- 4:55ADTK building blocks and API
- 7:25Basic outlier detection
- 9:58Quantile thresholding
- 12:53Temporal anomaly detection
- 15:42Double rolling aggregates
- 17:30Visualizing pipelines
- 21:30Level shifts and volatility
- 24:00Handling seasonality
- 27:15Autoregression for anomalies
- 29:24Multivariate techniques
- 31:35High-dimensional methods
- 33:44Custom detectors
- 35:32Evaluating models
- 38:48Tolerance and event expansion
- 40:13Time series splitting
- 41:44Real-world applications
related talks
21mCross-dataset Time Series Anomaly Detection for Cloud Systems
01h 16mAnomaly Detection
19mThe Rise of AI in Anomaly Detection
39mMastering real-time anomaly detection with open source tools
22mDavid Talby, SVP Engineering, Atigeo
30msktime: python toolbox for time series- next-generation AI
45mAnomaly Detection in Time Series
29mAnomaly Detection