A data scientist builds a feature preprocessing pipeline using scikit-learn and wants to log it as a single MLflow model so that preprocessing is applied correctly at inference time. Which approach is correct?

Question

Accepted Answer

B. Log the entire pipeline using `mlflow.sklearn.log_model(pipeline, artifact_path="model")`. The entire scikit-learn `Pipeline` object — including all preprocessing transformers and the final estimator — should be passed to `mlflow.sklearn.log_model()`. This ensures that the same preprocessing steps are automatically applied at inference time, preventing training/serving skew.

A data scientist builds a feature preprocessing pipeline using scikit-learn and wants to log it as a single MLflow model so that preprocessing is applied correctly at inference time. Which approach is correct?

Related Questions

Discussion