Skip to content
BENIBIX Learn

Training-Serving Skew

AI⚙️ MLOps & DeploymentProduction Monitoring

Also known as Train-Serve Skew, Training/Serving Skew, Online-Offline Skew

Live data and training data are made in two ways. So a good model does badly.

Think of it like

It is like practising football on grass. Then the real game is played on ice.

Example

Emma's model was trained on ages in years. The live app sends ages in months, so every customer looks 12 times older and the scores go wrong.

Why it matters

Nothing crashes, so it can hide for months. A model that tested well quietly gives poor answers.

Definition

Training-serving skew is when the data a model sees in real use differs from the data it was trained on, because of how the system was built. Often two pieces of code compute the same input in different ways. For example, training may use prices in dollars while the live app sends cents. The model still answers, but the answers are worse. It is a bug in the system, not a change in the world.
The fourth panel is training-serving skew. One input is built one way for training and another way live.

Common mistake

Calling it data drift. Drift is the world changing. Skew is your own two pipelines not matching.

Key takeaways

  1. 1
    Same input, two ways. Training and live code do not match.
  2. 2
    Silent harm. No error, just worse answers.
  3. 3
    Share the code. One feature step for training and serving.

Compare