フリー問題

問題文

Why would you package a feature retrieval specification together with the model artifact in Azure Machine Learning?

選択肢

1. So that the model can be deployed without a compute target
2. So that hyperparameter tuning can be rerun from the registered model, since the specification records which features were available at training time and the sweep needs that list in order to search over the same feature space again
3. So that inference uses exactly the same feature definitions and transformations that produced the training data, avoiding a mismatch between training and serving
4. So that the training data itself is copied into the model artifact