track clusters over time

[axiomofjoy]

A usability study participant wants to be able to track the performance on a cluster of data over time. The study participant saw a need to be able to check that, after exporting data from Phoenix and fine-tuning, his model performance had actually improved on similar production data points in the future.

This is an open problem. There is a temporal element here where we are looking at similar primary data points for different model versions. Interested to hear thoughts and ideas.