MLA-C01 Question 325
Select 3You are building a machine learning workflow to classify customer reviews into positive, neutral, or negative sentiments using Amazon SageMaker. After training a model, you notice that the model's performance metrics are inconsistent across different subsets of the test data, especially for reviews with technical jargon. What steps can you take to address the issue and improve the model's performance?
- A
Use Amazon SageMaker Data Wrangler to analyze the data distribution and identify imbalances in the dataset.
- B
Apply data augmentation techniques to increase the variety of training data, especially for reviews with technical jargon.
- C
Use Amazon SageMaker Hyperparameter Tuning to automatically optimize the model's hyperparameters for better performance.
- D
Switch to a different SageMaker instance type to increase the model's computational power.
- E
Implement stratified sampling to ensure the training dataset has a balanced representation of all sentiment categories.
Show answer and explanation
Correct answers: A, B, E
Explanation
The inconsistencies in model performance across subsets of the test data, particularly for reviews with technical jargon, suggest potential data imbalance or underrepresentation issues. Analyzing the data distribution using SageMaker Data Wrangler can help identify these issues. Data augmentation can enhance the diversity of the training dataset, and stratified sampling ensures balanced representation of all sentiment categories. These steps directly address the root cause of the issue, whereas hyperparameter tuning and switching instance types do not target the underlying data-related problem.
- A. Correct.
Analyzing the data distribution using SageMaker Data Wrangler can help identify imbalances or biases in the dataset that could affect the model's performance. This is a valid step to address inconsistencies in performance.
- B. Correct.
Data augmentation can improve the model's ability to generalize by exposing it to a wider variety of examples, especially for underrepresented cases like reviews with technical jargon. This is a valid step.
- C. Incorrect.
While hyperparameter tuning can improve model performance overall, it does not directly address inconsistencies in model performance caused by data issues such as imbalances or underrepresented subsets.
- D. Incorrect.
Switching to a different SageMaker instance type increases computational resources but does not address data quality or distribution issues that are likely causing the problem.
- E. Correct.
Stratified sampling ensures that all sentiment categories have balanced representation in the training data, which can help reduce performance inconsistencies across subsets. This is a valid step.