Google Professional Machine Learning Engineer Question 227
Select 4Google Cloud PlatformYou are tasked with training a machine learning model for a large-scale image classification task on Google Cloud. The dataset is stored in a Cloud Storage bucket and consists of millions of high-resolution images. Your goal is to optimize the training process to minimize cost and time while ensuring high model performance. Which considerations should you prioritize during the training process?
- A
Use preemptible VMs for training to save costs.
- B
Enable automatic hyperparameter tuning using AI Platform's Hyperparameter Tuning feature.
- C
Use a custom training loop instead of a prebuilt TensorFlow Estimator to reduce training time.
- D
Leverage Cloud TPU or GPU resources to accelerate large-scale training.
- E
Ensure the dataset is preprocessed and augmented before training begins.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
The correct answers focus on considerations that optimize the training process in terms of cost, time, and performance. Preemptible VMs reduce costs, while hyperparameter tuning improves model performance without extensive manual intervention. Leveraging Cloud TPU or GPU resources ensures efficient processing of large-scale datasets, and preprocessing the data ensures the model learns effectively from the input. Custom training loops are not necessary unless specific requirements demand them.
- A. Correct.
Using preemptible VMs is a cost-effective way to run training jobs, but you must ensure your training is designed to handle interruptions.
- B. Correct.
AI Platform's Hyperparameter Tuning feature can optimize model parameters, leading to better performance and potentially reducing the need for manual tuning.
- C. Incorrect.
While custom training loops offer flexibility, they do not inherently reduce training time. Prebuilt Estimators or frameworks like Keras may be sufficient for many use cases.
- D. Correct.
Using Cloud TPU or GPU resources can significantly accelerate training for large-scale datasets and high-complexity models.
- E. Correct.
Preprocessing and augmenting the dataset before training can improve model performance and ensure the data is in a format suitable for training.