Google Professional Machine Learning Engineer Question 229
Select 3Google Cloud PlatformYou are building a machine learning pipeline on Google Cloud to classify videos into categories. The dataset consists of thousands of videos stored in a Cloud Storage bucket. You need to efficiently organize the training data to ensure scalability and ease of access during training. What steps should you take to organize this video data?
- A
Store the videos in a Cloud Storage bucket using a directory structure that groups videos by category.
- B
Use a consistent and descriptive naming convention for the video files to encode metadata such as category or date.
- C
Convert the videos into tabular CSV files containing metadata and direct paths to the video files in Cloud Storage.
- D
Store the videos in BigQuery and use SQL queries to directly access the video data during training.
- E
Use Cloud Storage's Object Lifecycle Management to automatically delete videos after training to save costs.
Show answer and explanation
Correct answers: A, B, C
Explanation
To organize video training data effectively in Google Cloud, it is essential to use Cloud Storage to store the files with a structured directory hierarchy and consistent naming conventions. This ensures scalability and accessibility. Additionally, creating a tabular manifest file (e.g., a CSV) with metadata and file paths simplifies the data ingestion process for ML pipelines. While BigQuery and Object Lifecycle Management serve other purposes, they are not directly applicable to organizing raw training data for machine learning.
- A. Correct.
Storing videos in a Cloud Storage bucket with a directory structure based on categories improves organization and simplifies data loading for training pipelines.
- B. Correct.
Using a consistent naming convention helps encode useful metadata, making it easier to access and filter data during preprocessing and training.
- C. Correct.
Converting the dataset into a tabular CSV file with metadata and file paths is a common practice. This allows you to use the CSV file as a manifest to efficiently load and process videos from Cloud Storage.
- D. Incorrect.
BigQuery is not designed to directly store and process large binary files like videos. It is better suited for structured tabular data and metadata analysis.
- E. Incorrect.
Using Object Lifecycle Management to delete videos after training is not relevant to organizing training data. It is more about cost optimization and storage management post-training.