Google Professional Machine Learning Engineer Question 246
Select 4Google Cloud PlatformYou are training a machine learning model on Google Cloud AI Platform (Vertex AI). Your goal is to optimize the training process to minimize costs and training time while ensuring reproducibility. Which of the following actions are most appropriate to achieve your goal?
- A
Use preemptible VM instances for training jobs.
- B
Enable distributed training even if the dataset is small.
- C
Use managed datasets in Vertex AI to automatically split and preprocess data.
- D
Leverage hyperparameter tuning with a limited budget to find the best model configuration.
- E
Store training checkpoints regularly to resume training in case of interruptions.
Show answer and explanation
Correct answers: A, C, D, E
Explanation
To optimize the training process while minimizing costs and ensuring reproducibility, it is essential to use preemptible VMs for cost savings, manage datasets effectively to ensure reproducibility, leverage hyperparameter tuning with budget constraints to optimize the model, and store checkpoints to handle interruptions efficiently. Distributed training is only necessary for large datasets or complex models and is not appropriate in this scenario.
- A. Correct.
Using preemptible VM instances can significantly reduce costs for training jobs, but they may be interrupted. This is a cost-effective option for non-critical tasks or jobs with checkpointing enabled.
- B. Incorrect.
Distributed training is beneficial for large datasets or complex models, but it adds unnecessary overhead for small datasets and can increase costs and time. This is not recommended in this scenario.
- C. Correct.
Using managed datasets in Vertex AI simplifies preprocessing and splitting tasks, ensuring consistency and reproducibility. This is an efficient approach to handle your data.
- D. Correct.
Hyperparameter tuning with a limited budget allows you to optimize training without overspending. It is an effective way to find the best configurations while managing costs.
- E. Correct.
Storing training checkpoints regularly ensures that you can resume training in case of interruptions, which is particularly useful when using preemptible VM instances or when unexpected failures occur.