Exam Professional Machine Learning Engineer topic 1 question 280 discussion - ExamTopics

Problem

A Kubeflow pipeline for an end-to-end PyTorch-based MLOps workflow takes over an hour to complete. The pipeline involves reading data from BigQuery, processing data, feature engineering, model training, model evaluation, and deploying the model to Cloud Storage. The goal is to reduce pipeline execution time and cost.

Options

• A. Comment out parts of the pipeline not being updated.
• B. Enable caching in all pipeline steps.
• C. Delegate feature engineering to BigQuery and remove it from the pipeline.
• D. Add a GPU to the model training step.

Solution

The suggested answer is B. Enabling caching in all steps of the Kubeflow pipeline is the most efficient solution for speeding up execution and reducing cost without compromising the integrity of the process.