Technology selection

Based on many observations, drifting is built based on the best packages, that give it the easy start but also assure the best quality for the server and the used methods.

mlserver

mlserver is a modern, top-tier, feature-rich package for serving Machine Learning models. Building drifting on the top lets drifting use all the best design patterns.

Alibi Detect

Alibi detect implements online algorithms, uses methods proven to work by the research, and is very configurable.

Comparison to other packages

Alibi-detect-server from seldon-core

It is possible to build the solution in accordance with the framework proposed by drifting using seldon-core and alibi-detect-runtime.

The example of a similar project can be found here, where the drift detector can be chained to a model and can make the predictions always after the regular model prediction. However, this solution requires a lot of DevOps expertise, while in drifting, all the necessary steps can be easily done in Python, next to the regular model deployment.

Evidently

Evidently focuses on providing the tools for mathematical computation and visualization of drift. It's not a solution to detect the drift in production in the framework proposed by drifting.