Training-serving skew
Training-serving skew is a mismatch between the data, feature representations, or computations used during model training and those used to generate predictions.
Machine Learning System Design
A condition in which a machine learning model performs poorly because its training data does not represent the data encountered during application or serving.
- How it is used
- Frames production reliability as a data-representation problem requiring alignment between training inputs and application inputs.
- How deep it goes
- Explains causes, gives environmental and image examples, and recommends retraining or adaptation, without detailing implementation mechanics.
- Examples
- Uses crowdsourced versus real-world images, laboratory versus poor-lighting photos, and a St. Petersburg–Cyprus deployment mismatch.
- What you can do after
- Diagnose/debug with it by comparing production inputs with training data and responding through broader collection, labeling, retraining, or adaptation.
Read how Machine Learning System Design explains training-serving skew →
Machine Learning Platform Engineering
Training-serving skew is a discrepancy between the feature transformations or data seen during model training and those applied during inference.
- How it is used
- Prevents inconsistent features from entering training and inference workflows in production ML systems.
- How deep it goes
- Defines the discrepancy, identifies differing transformations as a cause, and stops short of detailed detection or remediation procedures.
- Examples
- Illustrates skew through different transformations during training and inference, including complex joins across multiple data sources.
- What you can do after
- Apply it via feature stores that serve transformed features through offline and online paths.
Read how Machine Learning Platform Engineering explains training-serving skew →
Related concepts
- feature storeaddressed by · A feature store is a system for managing and serving machine-learning features for use in model training and inference.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.