Pyod for Online and/or Incremental Learning

[xLolom]

Hi,

I would like to know if among all the algorithms implemented in PyOD, some of them are thought for online outliers detection, or for Incremental Learning, which are two important problematics in some applications of Anomaly Detection. (And if not, do you maybe know other open-source projects that have implemented such algorithms ?)

Thanks ! :)

[arnaudvl]

As far as I know, the implementations in PyOD are offline (please correct me if I'm wrong!), including for methods which could be online such as Loda. Anomaly, adversarial and drift detection library alibi-detect does have a couple of online methods such as an online version of the Mahalanobis distance for tabular data (extended to work on a mixture of numerical and categorical features) and Spectral Residuals for time series data.

[yzhao062]

This is correct. No online learning or streaming setting is considered for now.

[selimfirat]

In addition to the first answer by @arnaudvl, PySAD might be the one you are looking for. PySAD directly focuses on anomaly detection on streaming data. PySAD provides a complete set of tools for streaming experimentation and various models for online anomaly detection. It also provides wrappers to integrate PyOD models to the streaming setting, called Integrators.

PySAD Repository: https://github.com/selimfirat/pysad PySAD Documentation: https://pysad.readthedocs.io/en/latest/