liveBook · Manning

Share of drifted features

As explained in Machine Learning Platform Engineering

Share of drifted features is the proportion of monitored feature columns whose distributions are identified as having drifted.

Share of drifted features measures the proportion of monitored feature columns whose distributions are identified as drifted. In a real-time API monitor, reports are generated after the request window fills and then updated for each subsequent snapshot. A high value can reflect an overly small window or permissive thresholds rather than genuine production change. section 6.2.3

Interpreting a real-time snapshot

With a window size of 50 requests, a metric value near 0.875 indicates that approximately 87% of monitored columns drifted. Because the current data came from the training dataset, the result may be a sampling artifact; test larger windows and revise thresholds before triggering alerts. section 6.2.3

How Machine Learning Platform Engineering covers it

How deep it goes
Explains proportional interpretation and sampling effects, but stops short of defining the underlying statistical tests.
Examples
Uses 50-request windows, repeated API snapshots, an 0.875 metric value, and training data as the current-data reference.
What it assumes
Assumes monitored feature columns, reference data, incoming request windows, and configured report generation.
What you can do after
Diagnose/debug with it: distinguish possible sampling noise from actionable feature drift before changing thresholds or retraining models.
How it connects
Connects data drift with reference datasets, sliding windows, thresholds, drift alerts, and model retraining.
Cover of Machine Learning Platform Engineering
liveBookmanning.com
Read Machine Learning Platform Engineering on liveBook Get the book at manning.com

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Machine Learning Platform Engineering. The section links point to the parts of the book it draws on.