MLS-C01 Question 380
Select 3You have deployed a machine learning model using Amazon SageMaker, but predictions are consistently producing low accuracy compared to expected results. Which actions should you take to debug and troubleshoot the model?
- A
Check the input data for incorrect formatting or preprocessing errors.
- B
Analyze the feature importance to identify irrelevant or redundant features.
- C
Enable Amazon SageMaker Debugger to inspect model training metrics and identify potential issues.
- D
Increase the size of the training dataset without verifying data quality.
- E
Re-train the model with a higher number of epochs without investigating the current training process.
Show answer and explanation
Correct answers: A, B, C
Explanation
To debug and troubleshoot an ML model deployed on Amazon SageMaker, it is essential to investigate the data and feature preprocessing pipeline, use tools like Amazon SageMaker Debugger to analyze training metrics, and assess the relevance of features. Blindly increasing the training dataset size or re-training without diagnosing the underlying issue is not a recommended approach.
- A. Correct.
Checking the input data for formatting or preprocessing errors is a critical step in debugging ML models, as incorrect inputs can lead to poor predictions.
- B. Correct.
Analyzing feature importance helps identify irrelevant or redundant features that may negatively impact model performance, providing insights for feature engineering.
- C. Correct.
Amazon SageMaker Debugger allows you to inspect training metrics, such as loss and gradients, to identify potential issues during the model training process.
- D. Incorrect.
Increasing the size of the training dataset without verifying data quality may propagate errors in the data and does not address the root cause of the issue.
- E. Incorrect.
Re-training the model with more epochs without investigating the current training process may lead to overfitting or other issues without solving the accuracy problem.