DEA-C01 Question 391
Single answerYou are working on an AWS Glue ETL job to process a large dataset stored in Amazon S3 for machine learning model training. The dataset contains billions of records, and processing the entire dataset would be computationally expensive. You need to create a representative subset of the data to reduce costs while preserving the dataset's diversity. Which data sampling technique should you use to achieve this in AWS Glue?
- A
Simple random sampling
- B
Stratified sampling
- C
Systematic sampling
- D
Cluster sampling
Show answer and explanation
Correct answer: B
Explanation
Stratified sampling is the best choice in this scenario because it ensures that the sample is representative of the entire dataset by maintaining the proportional representation of different categories or strata. This is particularly important for machine learning datasets to prevent bias and ensure the diversity of the training data. In contrast, other sampling techniques like simple random sampling or systematic sampling may fail to capture the diversity of imbalanced datasets, and cluster sampling may result in underrepresentation of certain categories.
- A. Incorrect.
Simple random sampling selects data points randomly without considering specific categories or groups, which might not preserve diversity in a dataset with imbalanced classes.
- B. Correct.
Stratified sampling ensures that all groups or categories in the dataset are proportionally represented in the sample, making it suitable for preserving dataset diversity.
- C. Incorrect.
Systematic sampling selects data points at a regular interval, which can lead to biased results if the data has periodic patterns.
- D. Incorrect.
Cluster sampling selects entire groups or clusters, which may not preserve the diversity of the dataset if the clusters are not representative of the overall population.