MLflow Tracking and Registry

MLflow tracking records experiment context. The registry turns selected runs into governed model versions. Together, they help teams explain why a model was promoted.

Promotion evidence

• training data version,
• feature table versions,
• parameters and metrics,
• evaluation notebook or report,
• approval notes,
• deployment target.

Minimal tracking example

import mlflow

with mlflow.start_run():
    mlflow.log_param("features", "v3")
    mlflow.log_metric("validation_auc", 0.91)
    mlflow.sklearn.log_model(model, "model")

Review habit

Do not promote a model because it is newer. Promote it because the evidence shows it is better for the intended use.