search

yzhao062 / pytod

TOD: GPU-accelerated Outlier Detection via Tensor Operations
https://arxiv.org/abs/2110.14007
BSD 2-Clause "Simplified" License
169 stars 23 forks source link

Support for novelty detection and changing distance metric with local outlier factor #5

Open sugatc opened 1 year ago

sugatc commented 1 year ago

The current implementation of LOF doesn't allow changing the distance metric to 'cosine', for example or setting novelty = True which prevents it from being used for novelty detection task. It will be great if support can be added for these.

yzhao062 commented 1 year ago

thanks for the note. to my best knowledge, these distance metrics do not play a key role in detection. I have run quite a lot of benchmarks with different settings, and the results for kNN and LOF do not change quite a lot. So it might be fine to use euclidean.

sugatc commented 1 year ago

Thanks for your response. What about novelty detection -- I can see PyOd allows setting the parameter novelty to True. It will be useful to have this here since Pytod is much faster and can be used for large datasets. Is there a straightforward implementation of that?