tods/benchmark/synthetic/README.md

• Revisiting Time Series Outlier Detection:Definitions and Benchmarks
• • Resources
  • Cite this Work:
  • Datasets
  • Pipeline
  • Runner

Revisiting Time Series Outlier Detection:Definitions and Benchmarks

This branch is the source code of experiment part of our paper. We provide everything needed when running the experiments: Dataset, Dataset Generator, Pipeline json, Python script, runner and the result we get from the experiments.

Resources

• Paper: Under review

Cite this Work:

If you find this work useful, you may cite this work:

@misc{lai2020tods,
    title={TODS: An Automated Time Series Outlier Detection System},
    author={Kwei-Harng Lai and Daochen Zha and Guanchu Wang and Junjie Xu and Yue Zhao and Devesh Kumar and Yile Chen and Purav Zumkhawaka and Minyang Wan and Diego Martinez and Xia Hu},
    year={2020},
    eprint={2009.09822},
    archivePrefix={arXiv},
    primaryClass={cs.DB}
}

Datasets

This All of the datasets were generated from the generators. Tuning parameters in the generators can get your own datasets.
The datasets used in the paper is provided in "unidataset" and "multidataset"

Pipeline

This Pipeline json files are organized by different settings of algorithms. There are 5 outlier ratios in each setting.

Runner

To run a pipeline, you can generate your own pipeline json file from script. Take AutoEncoder as an example:

python script/simple_algo/build_AutoEncoder_pipeline.py 

Then run the json using run_pipeline in /runner

python runner/run_pipeline.py --pipeline_path ae_pipeline_default_con0.05.json 

Or you can also use the pipelines we provided in /Pipeline:

python runner/run_pipeline.py --pipeline_path Pipeline/AutoEncoder/ae_pipeline_default/ae_pipeline_default_con0.05.json

*Please refer master branch of TODS for details of running pipelines.