MLA-C01 Question 372
Single answerA machine learning engineer is deploying a model on AWS SageMaker for real-time inference. The model requires preprocessing of incoming data, such as normalization and encoding, before making predictions. The engineer needs to ensure that the preprocessing steps are applied consistently to both training data and inference requests. Which approach should the engineer take to achieve this?
- A
Use a SageMaker Processing job to preprocess the data during training and deploy a separate Lambda function to handle preprocessing during inference.
- B
Implement the preprocessing logic within a custom inference script and include it when deploying the SageMaker model endpoint.
- C
Preprocess the data manually for training and rely on SageMaker to automatically handle preprocessing during inference.
- D
Use SageMaker's built-in feature transformation options to preprocess the data during training and inference.
Show answer and explanation
Correct answer: B
Explanation
When deploying a model for real-time inference, it is critical to apply the same preprocessing logic to both the training data and inference requests to ensure consistency and accurate predictions. By integrating preprocessing logic into a custom inference script, the engineer can guarantee that the same steps are applied during both phases, avoiding potential mismatches or inconsistencies.
- A. Incorrect.
Using a SageMaker Processing job and a Lambda function separates training and inference preprocessing steps, which could lead to inconsistencies if the logic is not identical between the two.
- B. Correct.
Implementing preprocessing logic in a custom inference script ensures that the same preprocessing steps are consistently applied during both training and inference. This approach is highly recommended for custom requirements.
- C. Incorrect.
Manually preprocessing data during training and relying on SageMaker for automatic preprocessing during inference creates a potential mismatch in preprocessing logic, as SageMaker does not automatically handle custom preprocessing.
- D. Incorrect.
SageMaker's built-in feature transformation options are limited to specific use cases. Custom preprocessing steps, such as normalization and encoding, may not be supported natively.