Google Professional Machine Learning Engineer Question 252
Single answerGoogle Cloud PlatformYour team is building a machine learning model for predicting customer churn. You need to train the model on a large dataset stored in Google Cloud Storage. The team has decided to use a TensorFlow-based training script. The training process needs to be scalable and take advantage of Google Cloud's managed services. Which approach should you take to best meet these requirements?
- A
Use Vertex AI custom training to submit the TensorFlow training script as a Python package and configure the required compute resources.
- B
Manually set up a Google Kubernetes Engine (GKE) cluster and run the TensorFlow training script using Kubeflow pipelines.
- C
Run the TensorFlow training script directly on a Compute Engine instance with pre-configured GPUs.
- D
Use Vertex AI pre-built algorithms and skip custom training.
Show answer and explanation
Correct answer: A
Explanation
Vertex AI custom training is the best choice because it allows you to submit a TensorFlow training script, configure compute resources like GPUs or TPUs, and leverage Google Cloud's managed services for scalable training. It eliminates the need for manual infrastructure setup and management, which is required for Kubeflow on GKE. Additionally, running the script on a Compute Engine instance may not meet the scalability requirements, and pre-built algorithms cannot handle custom scripts.
- A. Correct.
This option is correct because Vertex AI custom training allows you to easily submit TensorFlow training scripts as Python packages, configure compute resources, and manage the training process in a scalable and managed way.
- B. Incorrect.
This option is not ideal because setting up and managing a GKE cluster with Kubeflow requires significant manual effort and expertise, whereas Vertex AI custom training provides a simpler, managed solution for this scenario.
- C. Incorrect.
Running the training script on a single Compute Engine instance does not provide the scalability required for training on a large dataset, and managing resources manually is less efficient than using Vertex AI.
- D. Incorrect.
While Vertex AI pre-built algorithms can be useful, they are not applicable here because the team is using a custom TensorFlow training script, which requires a custom training solution.