Implementation of the StandardAbsoluteDeviation (SAD) anomaly detection algorithm

[hoanganhngo610]

This PR contains the implementation of the Standard Absolute Deviation (SAD) anomaly detection algorithm.

This implementation is adapted from the implementation within PySAD (Python Streaming Anomaly Detection) package, which calculates the anomaly score by dividing the deviation from the mean/median to the standard deviation of all points seen within the data stream. This idea is adapted from the paper by Hochenbaum et al. (2017).

Despite being a fairly simple algorithm, the implementation of this algorithm allows the anomaly submodule of River to have a solid competitor to conduct any benchmarking work.

There is no unit test for this algorithm since the result has already been compared locally with the implementation in PySAD, which yielded exactly similar results. Creating unit tests upstream would require the installation of another package, which seem unnecessary.

[hoanganhngo610]

@MaxHalford Would you mind having a look through this PR? There are two problems that are making the tests fail, including

• mypy: I believe these errors do not come from the files that I modified.
• test_estimators tests: I believe I have added SAD to the list of estimators that are expected to be ignored upon testing, but somehow they are still tested. I personally do not encounter these errors running pytest locally.

Thank you so much in advance!

[MaxHalford]

The issue with this method is that it is univariate. You're trying to fit into the multivariate anomaly detection framework, which is not the right way.

What you should do is inspire yourself from anomaly.GaussianScorer. It's also a univariate anomaly detection method.

Let me know if this isn't clear

[hoanganhngo610]

@MaxHalford I think I can understand the idea, and have also implemented it in my upcoming commits. It would be much appreciated if you can have a look through and see whether the new approach is appropriate!

[MaxHalford]

Cool! Can you add a note to unreleased.md?

[hoanganhngo610]

@MaxHalford Thank you so much for approving the changes! The entry has also been added to the UNRELEASED.md file.