Question

You train a scikit-learn classifier in an Azure Machine Learning notebook and use MLflow autologging. After the run completes, you want to register the resulting MLflow model so that consumers get an inferred model signature (input/output schema) and an input example stored with the artifact for easier validation at deployment time. What should you do during training to ensure the signature and input example are captured before registering the model?

Accepted Answer

Call mlflow.sklearn.log_model() explicitly, passing the signature (inferred with mlflow.models.infer_signature) and an input_example, then register that logged model.

Answer

Register the model from the run's autologged artifact directly, since autologging always stores a signature and input example by default.

Answer

Convert the model to ONNX format after training, because only ONNX models support a stored signature in the model registry.

Answer

Add the signature manually by editing the MLmodel file in the registered model's storage container after registration.