Answer: Retrain the model when a significant shift in the distribution of customer attributes is detected in the production data compared to the training data.

The question emphasizes minimizing both cost and effort in retraining a customer churn prediction model on Vertex AI, with new training data available weekly. Option A is optimal because it triggers retraining only when a significant shift in the distribution of customer attributes (i.e., data drift) is detected in production data compared to training data. This approach avoids unnecessary retraining, reducing computational costs and manual effort, while ensuring the model adapts to meaningful changes that impact performance. Option C (retraining when accuracy drops by 10% on new data) is less suitable because it requires frequent evaluation on new datasets, increasing effort and cost, and may not align with real-world performance shifts. Option B (retraining due to latency increases) is irrelevant, as latency is a deployment/infrastructure issue, not a model performance one. Option D (weekly retraining) is inefficient, as it retrains regardless of need, leading to high costs and effort without guaranteed benefits. The community discussion shows a split between A and C, but A is favored for cost minimization, as it retrains 'only when needed'.

Author: LeetQuiz Editorial Team