Google Professional Machine Learning Engineer Question 100
Select 3Google Cloud PlatformYou are building a machine learning pipeline on Google Cloud to process and train a model using a diverse dataset consisting of tabular data, images, and text. Which of the following steps should you take to properly organize and prepare the data for training?
- A
Store tabular data in BigQuery and use SQL queries to preprocess it.
- B
Store images in Cloud Storage and use a structured naming convention to organize them.
- C
Store text data in Cloud SQL for easy access and query processing.
- D
Convert all data types into a single CSV file for uniformity before training.
- E
Use Dataflow pipelines to preprocess and transform data from different sources.
Show answer and explanation
Correct answers: A, B, E
Explanation
To organize and prepare diverse datasets for machine learning on Google Cloud, use tools suited to each data type's characteristics. BigQuery is ideal for tabular data, Cloud Storage works well for unstructured data like images, and Dataflow enables preprocessing and transformation across data sources. Ensuring data is properly stored and preprocessed is crucial for effective model training.
- A. Correct.
Storing tabular data in BigQuery is efficient for handling large-scale structured data, and SQL queries provide powerful tools for preprocessing before training.
- B. Correct.
Cloud Storage is a scalable and cost-effective solution for storing images, and a structured naming convention ensures the data is organized and easy to retrieve.
- C. Incorrect.
Cloud SQL is designed for relational databases and may not be the best choice for storing large-scale unstructured text data, especially when alternatives like Cloud Storage or BigQuery are more appropriate.
- D. Incorrect.
Converting all data into a single CSV file is not a scalable or efficient solution for datasets containing multiple modalities (e.g., images, text, tabular data).
- E. Correct.
Dataflow pipelines allow you to preprocess and transform data from multiple sources (e.g., BigQuery, Cloud Storage) in a scalable and automated manner, making it ideal for preparing diverse datasets for training.