TSAD-Forge
A reproducible benchmarking and learning ecosystem for time-series anomaly detection (TSAD), organized by generation of techniques and evaluated under one honest protocol.
git clone https://github.com/Denny-Hwang/TSAD-Forge.git && cd TSAD-Forge
pip install -e ".[dev]"
tsad-forge run --model dummy --data synthetic
- Learn — theory track ch01–ch10 with companion notebooks
- Leaderboard — VUS-PR primary metric, 8 interactive charts
- Datasets — dataset cards: sources, licenses, known flaws
Generations
| Generation | Era | Representative methods |
|---|---|---|
| Gen1 Statistical | 1930s–2000s | CUSUM, EWMA, Hotelling T², PCA-T²/SPE, Sub-PCA, STL, POLY |
| Gen2 Classical ML | 2000–2016 | LOF, OC-SVM, IForest, KNN, Sub-KNN, Matrix Profile |
| Gen3 DL Recon/Forecast | 2015–2020 | AE, LSTM-AD/P, VAE (Donut), DAGMM, OmniAnomaly, USAD |
| Gen4 Graph/Transformer | 2020–2023 | GDN, MTAD-GAT, Anomaly Transformer, TranAD, DCdetector, TimesNet |
| Gen5 SSM/Foundation | 2023– | MambaTSAD (faithful/fixed), MOMENT, Chronos, TimesFM 2.5 / 3.0 (multivariate · channel-independent · probabilistic) |
Evaluation methodology comes first
Point adjustment (PA) makes even random scores look state-of-the-art
(Kim et al., AAAI 2022). This repository uses VUS-PR as the primary metric;
PA-F1 is only available behind a --legacy-pa flag with a warning attached.
Reproduce the inflation yourself in ch07.