search

IRC-SPHERE / HyperStreamOnlineLearning

Plugins for Online Learning (supervised or unsupervised) in HyperStream
MIT License
3 stars 1 forks source link

Add unsupervised methods #2

Open perellonieto opened 7 years ago

perellonieto commented 7 years ago

Some easy to add tools could be density estimation models that learn the distribution of the data without any labels.

Also, one class classification models, novelty or anomaly detection algorithms.

tdiethe commented 6 years ago

As a general note, possibly you could support the base classes in sklearn (ClassifierMixin, RegressorMixin) rather than individual methods, and then the actual methods could be passed as a parameter (possibly a string, although that's ugly) which would mean less work on your end.

perellonieto commented 6 years ago

Commit 3260fe6518734fc3be6921ea0324510c179d59dd adds a tool for anomaly detection. That is, it uses a model that can be trained in an unsupervised way and generates scores at every stream. Because the model is passed as a parameter, it is not only anomaly detection but could be density estimation, outlier detection, clustering, or any other unsupervised method. The tool should be refactored in order to indicate that.

perellonieto commented 6 years ago

Open #3 to refactor the anomaly detection tool into unsupervised learning.