EngineersofAI LogoCoursesBlogResearch LabAI LettersThe LabCode BankInteractive 3DKodrEarnestJobs
Skip to main content
Interactive 3D/Data and Concept Drift
PSI0.0000
0.1
0.2
No drift
PSI
0.0000
KL Divergence
0.0000
Month
0 / 18
Prod Mean
40.0
Train Mean
40.0
User Age: Gradual drift - user base is aging slowly. PSI rises steadily.
Recommended action: No action needed - monitor weekly.
Drift Controls
Feature
018
0.10.3
Data drift = input distribution changes (users' behavior changed).

Concept drift = the relationship between inputs and outputs changes (the world changed).

Both degrade model performance without triggering any code errors.

PSI > 0.2 is the industry standard threshold for triggering retraining.

Data and Concept Drift - Interactive Visualization

Models degrade silently in production when the real world changes. Data drift means the input distribution shifts (users changed their behavior); concept drift means the relationship between inputs and outputs changes (the world changed). Both cause model accuracy to drop without triggering any error - unless you monitor for them. PSI (Population Stability Index) is the standard industrial metric.

  • Gradual drift: User Age feature slowly shifts over 18 months - see PSI creep past 0.1 threshold
  • Sudden drift: Request Length feature changes abruptly at month 6 - PSI jumps to 0.3+
  • Concept drift: output distribution changes even when inputs look the same
  • PSI gauge: green (< 0.1 no action), orange (0.1-0.2 investigate), red (> 0.2 retrain)
  • KL divergence shown alongside PSI - see which metric detects drift faster
  • Used in production ML monitoring: Evidently, WhyLogs, Arize, Fiddler, SageMaker Monitor

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.