Reason for using `find_epsilon` in feature-level

ML4ITS / mtad-gat-pytorch

PyTorch implementation of MTAD-GAT (Multivariate Time-Series Anomaly Detection via Graph Attention Networks) by Zhao et. al (2020, https://arxiv.org/abs/2009.02040).

MIT License

328 stars 76 forks source link

Reason for using `find_epsilon` in feature-level #14

Closed cloudhs7 closed 2 years ago

cloudhs7 commented 2 years ago

Hi. Thanks for your excellent work! I'm wondering why only epsilon method is used in predicting anomalies at feature-level in below.

https://github.com/ML4ITS/mtad-gat-pytorch/blob/9e671ea99dedd82ac55f53e53af1d1b56c13ebff/prediction.py#L142

Can POT method also be used in here, or is there any reason for using only epsilon method in here? Thanks for your help.

axeloh commented 2 years ago

Hello,

There is no particular reason for why exactly (and only) the epsilon method is used for this. The feature-wise thresholds can be used for diagnosis and visual purposes, which is why we include it. However, for actually detecting anomalies, we care only about the threshold at entity-level (feature-wise aggregated), where both epsilon and POT are included for comparison.