Analyze historical data drift
TL;DR: You can use Evidently to evaluate drift in your training data. Understanding past data changes helps define monitoring and retraining strategy. Here is a sample Jupyter notebook.

Tutorial - "How to detect, evaluate and visualize historical drifts"

In this tutorial, we check historical drift on a month-by-month basis. We add a custom function to set drift conditions on the dataset level. We use Evidently Data Drift Profile to calculate drift, Plotly to generate custom visuals, and integrate with Mlflow to log the results.
Last modified 2mo ago