The repository for the paper CLEDAD: Contrastive Learning Enhanced Conditional Diffusion for Time Series Anomaly Detection.
This is a refactored version of the resulting code of the paper for ease of use. Follow these steps to copy each cell in the result table.
This code needs Python-3.8 or higher.
conda create --name cledad python=3.8 -y
conda activate cledad
pip3 install -r requirements.txt
Download public datasets used in our experiments:
python src/utils/download_data.py [dataset-name]
Options of [dataset-name]: msl, smd, smap, swat and psm.
To run the experiments on different dataset, you can just run the following command:
python src/experiments/train_test_anomaly_detection.py --dataset [dataset-name] --device [device] --seed [seed] --anomaly_ratio [anomaly_ratio] --config [config-name]
The values of [dataset-name], [seed] and [anomaly_ratio] used in our experiments are available in our paper.
You can modify the config file to train and test with different parameters. The config file is located in src/config directory. For different datasets, we used different [dataset-name].yaml configurations
The code for this library is referenced from the following repositories, in particular data download and processing: