A fundamental problem faces with network traffic is that the traffic is dynamic and evolves over time. Presented herein is a tool for monitoring feature distribution shift and possible efficacy implications. The tool takes advantage of the fact that labels are available on the training set, but not available in the production environment where the labelling cannot be done on the fly. The tool implicitly provides estimates of feature robustness to time shift for each feature.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Brabec, Jan; Machlica, Lukas; and Skarda, Cenek, "MONITORING OF FEATURE-DISTRIBUTION SHIFT IN PRODUCTION ENVIRONMENT", Technical Disclosure Commons, (November 29, 2018)