TSDynamicer

Time Series Dynamic Anomaly Detection with Transformer

Dependencies

python >= 3.6

pip3 install -r requirements.txt

Data Preparation

single index dataset with timestamp

AIOps challenge 2018: download and unzip data.

Web of AIOps

AIOps-github

NAB: git download.

NAB

multi index dataset with timestamp

SkAB: git download.

SkAB

custom dataset

Custom Server Metrics (CSM): non-public dataset.

Get Started

all datasets will be processed to dataframe with the same sample_obj list: [sample_obj1, sample_obj2, ...],details:

sample_obj.sample_time: current time string

sample_obj.dataset: dataset name

sample_obj.data_des: which column or file is used to get data

sample_obj.sample_data: processed target data of current sample, processed sample df

sample_obj.label: label of current sample, 0 for exception and 1 for normal

start script (e.g. SkAB)

bash ./scripts/SkAB/SkAB-TSDynamicer.sh

Further Reading

Citations

Contact

If you have any question or want to use the code, please contact [email protected].

Acknowledgement

We appreciate the following github repositories:

https://github.com/zhouhaoyi/Informer2020

https://github.com/thuml/Autoformer

https://github.com/MAZiqing/FEDformer

https://github.com/cure-lab/LTSF-Linear

https://github.com/thuml/Nonstationary_Transformers