A data scientist trains a scikit-learn StandardScaler on the training split and saves it alongside the trained model weights. Which term best describes the StandardScaler object in the context of ML artefact management?
- AA preprocessing artefact Correct
- BA hyperparameter configuration file
- CA dataset version snapshot
- DAn evaluation metrics record
Why A is correct: A fitted preprocessing transformer such as StandardScaler is a preprocessing artefact: a serialisable object produced during the training pipeline that must be versioned and stored alongside the model so that inference inputs can be transformed identically to training inputs.
Why B is wrong: Hyperparameter files record settings such as learning rate or max depth, not fitted transformation state. A StandardScaler holds computed mean and variance values that are required to reproduce the exact transformation applied at training time.
Why C is wrong: A dataset version snapshot refers to a recorded, immutable cut of raw or processed data, not a fitted transformation object. Conflating the two would cause an artefact store to mis-categorise the object and break reproducibility checks.
Why D is wrong: Evaluation metrics records store scalar or aggregate performance values such as accuracy or RMSE. A StandardScaler contains no performance measurement; it holds statistical parameters derived from training data and applied during feature transformation.