liveBook · Manning

Data drift report

As explained in Machine Learning Platform Engineering

A data drift report is a generated analysis that compares feature distributions in reference and current datasets and reports statistical test results.

A data drift report compares each feature's reference and current distributions, identifies the statistical test used, and records its drift score or p-value. With Evidently, DataDriftPreset generates a default report across columns and can expose drifted-column counts and missing values. section 6.2.1

A production workflow typically follows these steps:

  • Interpret Wasserstein distance scores against Evidently's default 0.1 drift threshold. section 6.2.1

Minimal Evidently report

report = Report(metrics=[DataDriftPreset()]) report.run(reference_data=reference_data, current_data=current_data) report.save_html("drift_report.html")

How Machine Learning Platform Engineering covers it

How deep it goes
Explains report generation, feature comparisons, Wasserstein scoring, and thresholding, but defers longitudinal performance tracking.
Examples
Uses DataDriftPreset, Wasserstein distance, HTML export, daily or API snapshots, and a drift dashboard.
What it assumes
Assumes reference and current data frames; teaches neither statistical testing foundations nor snapshot acquisition.
What you can do after
Operate/configure an Evidently report that evaluates feature drift and saves the result for inspection.
How it connects
Connects statistical tests, reference datasets, current datasets, data quality reports, regression reports, and dashboards.
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.