MLA-C01 ML Model Development Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✓

  Configure a built-in rule for vanishing gradients, such as the VanishingGradient rule.

  Why this is correct

  SageMaker Debugger provides built-in rules like VanishingGradient that analyze tensor outputs during training to detect when gradients become very small. This rule can trigger an alert or stop the training job if vanishing gradients are detected, allowing the engineer to address the issue early. It is specifically designed for this purpose and requires minimal configuration.

• ✗

  Use SageMaker Clarify to detect bias in the gradients.

  Why it's wrong here

  SageMaker Clarify is designed for bias detection and explainability in data and models, not for monitoring gradients during training. It does not analyze training tensors or loss curves. Using Clarify would not help detect vanishing gradients or loss plateaus; it is a separate service for fairness and interpretability.

• ✗

  Set up a custom rule using a Python script that analyzes gradients and loss.

  Why it's wrong here

  While custom rules are possible, they require additional development effort and are not necessary when built-in rules already exist for the specific issues. The engineer wants to achieve the goal efficiently, so using built-in rules is preferable. Custom rules might be used for more complex or unique conditions, but here the built-in rules suffice.

• ✓

  Enable Debugger's LossNotDecreasing rule to monitor the loss curve.

  Why this is correct

  The LossNotDecreasing built-in rule monitors the loss value over time and triggers if the loss does not decrease according to a specified threshold. This directly addresses the requirement to detect when the loss is not decreasing. It is a standard rule provided by SageMaker Debugger and can be easily enabled in the estimator configuration.

• ✗

  Configure the training job to save all tensors to Amazon S3 for manual analysis.

  Why it's wrong here

  Saving all tensors to S3 for manual analysis is inefficient and does not provide real-time alerts. The engineer wants to detect issues early, which requires automated rules that run during training. Manual analysis would be after-the-fact and would not prevent failure. This approach also incurs additional storage costs and latency.

Go deeper

Related to this question