สร้างรายงานการเคลื่อนด้วย Evidently

Raw PSI and KS numbers are great for code, but teammates want a clear picture. Evidently turns drift checks into shareable visual reports. 📈

Evidently is an open-source Python library that compares a reference dataset against current data and renders ready-made drift and quality reports.

Getting started is a single pip install. Then you import the report and preset you need for the analysis.

Every Evidently report needs two frames: a reference baseline, usually training data, and the current production data you want to compare against it.

A preset bundles many checks at once. DataDriftPreset runs a per-feature drift test and summarizes how many features moved.

Build a Report with your chosen metrics, then run it on both datasets. Evidently picks sensible drift tests per column automatically.

Export the report to a single HTML file. Anyone can open it in a browser, no Python required, which makes sharing painless.

For automation, pull the report out as a dict instead of HTML. Now your pipeline can read drift flags and act on them in code.

Evidently is not only for drift. The DataQualityPreset flags missing values and odd ranges, so you catch broken inputs alongside shifting ones.

When you want a clear verdict, use an Evidently TestSuite. It returns pass or fail per check, perfect for gating a CI pipeline.

The real power comes from running reports on a schedule, like daily, and storing each one. Over time you build a living history of your data health. ⏱️

One question on wiring drift into automation.

You met Evidently: feed it reference and current data, run a preset, then save HTML for humans or as_dict for automation. Schedule it to track data health over time. 🎯