My team deals with lots of time series data and in particular we are faced with anomaly detection problems on time series. To help us deal with that more efficiently, we built a toolkit in Python, ADTK, to quickly and easily test out different anomaly detection models and data flows.

ADTK has an API that allows you to easily combine a large number of anomaly detection models ("detectors"), data transformers, and ensembling steps ("aggregators") into serial or parallel data flows ("pipelines" and "pipenets"). It can also be easily extended.

We've just recently released ADTK under an open source license (MPL). We'd love to have people try it out, make contributions, and ask any questions.

ADTK is pip installable and the documentation and examples are linked in the readme on Github.

Most of the credit goes to Tailai Wen, who led this effort.

Here is a blog article I wrote introducing the package. https://www.arundo.com/arundo_tech_blog/adtk-open-source-tim...

This is awesome, thanks for sharing. It feels like this could be useful for detecting data quality issues in things like daily ETL, I'll have to give it a try.

Thanks. This is really comprehensive and well documented.