Change threshold logic in explainer
- once most relevant features are determined for local explanations, compute univariate ECDFs for each continuous feature (based on org. scales).
Then compute the empirical (1-p)% confidence interval of the observations.
If an observation is not an element of that interval consider it as relevant (w.r.t. anomaly score expl.)
- Readme.md change