Google Professional Machine Learning Engineer Question 245
Select 3Google Cloud PlatformYou are training a machine learning model on Google Cloud AI Platform using a large dataset stored in BigQuery. Your model training job is taking much longer than expected. After reviewing the setup, you identify that the training process is bottlenecked by data input. Which of the following steps should you take to optimize the training process?
- A
Enable BigQuery Storage API to speed up data retrieval.
- B
Switch to using a smaller training dataset to reduce input size.
- C
Use TensorFlow Data Pipelines to preprocess and stream the data efficiently.
- D
Increase the machine type to a more powerful one, such as an NVIDIA A100 GPU.
- E
Use Cloud Dataflow to preprocess the data and store it in Cloud Storage for faster access during training.
Show answer and explanation
Correct answers: A, C, E
Explanation
The training process is bottlenecked by data input, so optimizing the data pipeline is key. The BigQuery Storage API, TensorFlow Data Pipelines, and Cloud Dataflow are all effective solutions for improving data input throughput. Simply increasing machine power or reducing dataset size does not directly address input bottlenecks or could negatively impact model performance.
- A. Correct.
The BigQuery Storage API allows faster data retrieval compared to standard queries, which can help reduce the data input bottleneck.
- B. Incorrect.
Switching to a smaller dataset would reduce data input size but compromises the model’s ability to generalize, which is not an optimal solution.
- C. Correct.
TensorFlow Data Pipelines are designed to handle large datasets efficiently by preprocessing and streaming data, which helps prevent input bottlenecks.
- D. Incorrect.
Increasing the machine type may improve training speed but does not directly address the data input bottleneck.
- E. Correct.
Cloud Dataflow can preprocess large datasets and store them in Cloud Storage, which is optimized for high-throughput access during training.