search

yzhao062 / DCSO

Supplementary material for KDD 2018 workshop "DCSO: Dynamic Combination of Detector Scores for Outlier Ensembles"
https://www.andrew.cmu.edu/user/lakoglu/odd/accepted_papers/ODD_v50_paper_3.pdf
19 stars 7 forks source link

Any Expected Release Date of the Production version? #1

Open fcoppey opened 5 years ago

fcoppey commented 5 years ago

Hi,

I work on NIR spectra and I've been using your PyOD codes for my research in order to determine whether a new multivariate observation was similar enough to my known observations before allowing my quantification models to be applied on the new observation and it worked really great. now my dataset has become bigger and i have realized that i could only reach a good quantification with an ensemble models. I am also getting more and more false positives with using a unique KNN models for outliers detection. I am therefore wondering if the version you will provide here will help me gerenerate an ensemble predictor of one-class algorithms? or is it already possible to do so with the PyOD libraries?

Thank you very much for your amazing work. so helpful!

yzhao062 commented 5 years ago

Thanks for the interest:)

The production version is going to be released in pyod (aiming for another week or two): https://github.com/yzhao062/pyod/tree/lscp_implementation

The improved version of DCSO is called LSCP (https://github.com/yzhao062/LSCP).

However, I would recommend using IForest or HBOS in pyod for high dimensional data for now. Have you tried it? I feel Isolation Forest might be a good case here. https://github.com/yzhao062/pyod/blob/master/pyod/models/iforest.py