Training-serving skew

How it is used

Used in model deployment reviews. Shared code can help, but timing, missing values and data sources also need comparison.

Example sentences

1. We found training-serving skew because the live service filled missing ages with zero.

   The live feature preparation differed from the rule used to prepare training examples.

2. We compared stored training features with online values to investigate training-serving skew.

   Comparing the two paths can reveal differences hidden by good offline evaluation scores.

Common collocations

• detect training-serving skew
• reduce training-serving skew

Related terms

• MLOpsThe practices and tools for running machine learning reliably in production: versioning data and models, automating training and deployment, and monitoring models over time.
• preprocessingThe steps that clean and prepare raw data before a model uses it, such as removing duplicates, filling in missing values, resizing images or normalizing text.

More terms: Machine Learning and Data Processing

• artificial intelligence
• machine learning
• model
• dataset
• training data
• label
• train
• supervised learning
• neural network
• deep learning
• loss function
• overfitting