Google Professional Machine Learning Engineer
Get started today
Ultimate access to all questions.
You work for an online publisher that delivers news articles to over 50 million readers. As part of your role, you have built an AI model to recommend content for the company’s weekly newsletter. The success of a recommendation is determined if the recommended article is opened within two days of the newsletter’s publication date and the user spends at least one minute on the page. The data required to calculate this success metric is stored in BigQuery and is updated on an hourly basis. The AI model, trained on eight weeks of data, typically sees its performance drop below an acceptable baseline after five weeks, and retraining the model takes approximately 12 hours. Your goal is to ensure the model maintains performance above the acceptable baseline while also minimizing costs. How should you monitor the model to determine when it needs to be retrained?
You work for an online publisher that delivers news articles to over 50 million readers. As part of your role, you have built an AI model to recommend content for the company’s weekly newsletter. The success of a recommendation is determined if the recommended article is opened within two days of the newsletter’s publication date and the user spends at least one minute on the page. The data required to calculate this success metric is stored in BigQuery and is updated on an hourly basis. The AI model, trained on eight weeks of data, typically sees its performance drop below an acceptable baseline after five weeks, and retraining the model takes approximately 12 hours. Your goal is to ensure the model maintains performance above the acceptable baseline while also minimizing costs. How should you monitor the model to determine when it needs to be retrained?
Explanation:
Option C is the best answer. Scheduling a weekly query in BigQuery is a cost-effective and straightforward way to monitor the model's performance since all the necessary data to compute the success metric is already available and updated hourly in BigQuery. By conducting a weekly check, you can determine if the model’s performance has declined below the acceptable baseline and decide if retraining is necessary, avoiding unnecessary retraining and additional monitoring infrastructure costs.