EMMV_benchmarks

How to evaluate unsupervised Anomaly Detection algorithms?

author: Nicolas Goix, goix.nicolas@gmail.com

This is the code associated with ICML workshop paper https://arxiv.org/abs/1607.01152

-File em_bench.py evaluates the algorithms using EM and MV based criteria, without sub-sampling features nor averaging. It does not work in high dimensions.

-File em_bench_high.py makes use of sub-sampling (along features) and averaging, extending the use of these criteria to high-dimensional datasets.

-Basic EM and MV calculation is implemented in em.py