Support for KDE - Githubissues

yzhao062 / pyod

A Python Library for Outlier and Anomaly Detection, Integrating Classical and Deep Learning Techniques

http://pyod.readthedocs.io

BSD 2-Clause "Simplified" License

8.49k stars 1.36k forks source link

Support for KDE #301

Open hanshupe opened 3 years ago

hanshupe commented 3 years ago

Is there any support for KDE (Kernel Density Estimation) based outlier detection planned?

yzhao062 commented 3 years ago

I think this is a classical approach but have not been implemented in PyOD yet. before it is actually implemented, you may use HBOS which is sort of non-parameteric density estimation...much simpler tho

psmgeelen commented 3 years ago

How would that work? I have a little experience with KDEpy, which is a fast library for doing KDE. Do you want to measure the density-score of a given point based on the Density Estimation? I might have something I can share with you for that :)

tam17aki commented 3 years ago

Hi,

I have tried to implement KDE using pyod and scikit-learn:

class definition https://gist.github.com/tam17aki/480a2c3155e1e06d59231c8f5eac65f4
notebook for example https://gist.github.com/tam17aki/8bd3b6f5fe0091c29f61358a2cc00f6e

I hope this code will help those who try to incorpolate it into the official pyod package.

Best regards, Akira Tamamori