Skip to content

Latest commit

 

History

History
30 lines (22 loc) · 1.01 KB

README.md

File metadata and controls

30 lines (22 loc) · 1.01 KB

THOC-Pytorch

(Unoffical) Implementation of THOC model : Lifeng Shen, Zhuocong Li, James T. Kwok: Timeseries Anomaly Detection using Temporal Hierarchical One-Class Network. NeurIPS 2020.

If you have noticed errors in implementing, or found better hyperparamters/scores, plz let me know via github issues, pull request, or whatever communication tools you'd prefer.

Quickstart

touch secret.py < echo "WANDB_API_KEY={your_wandb_api_key}" # this repo utilizes wandb.
sh script.sh {gpu1} {gpu2} {gpu3} {gpu4}

Wandb Sweep

wandb sweep hptune/{data_name}.yaml
CUDA_VISIBLE_DEVICES={gpu_id} wandb agent {sweep_id}

Re-implementation so far

THOC uses F1-PA metrics for evaluation:

Paper Our Repo
NeurIPS-TS-MUL - -
SWaT - -
MSL - -
SMAP - -