03_model_registration_and_validation.md¶

Model Registration and Validation¶

Once the best model is identified, it must be formally registered in the MLflow Model Registry to manage its lifecycle.

1. Selecting the Best Run¶

The register_model component: - Queries the MLflow tracking server for all runs in the experiment. - Sorts the runs by rmse in ascending order. - Selects the run with the absolute minimum RMSE.

2. Model Registration¶

The model from the best run is registered as BikeSharingModel_pipeline_{PARTICIPANT_FIRSTNAME}. This assigns a version number to the model, allowing for easy rollbacks and version tracking.

3. Post-Registration Validation¶

To ensure the registered model is functional, the component performs a "sanity check": - Loads the registered model using mlflow.pyfunc.load_model. - Passes a sample input DataFrame. - Verifies that the prediction is a positive number.

This step prevents "broken" models from progressing to the deployment stage.