A company has 2,000 retail stores. The company needs to develop a new model to predict demand based on holidays and weather conditions. The model must predict demand in each geographic area where…

Question

Options

• ADevelop the model by using the Amazon Forecast Prophet model.
• BDevelop the model by using the Amazon Forecast holidays featurization and weather index.
• CDeploy the model by using a canary strategy that uses Amazon SageMaker and AWS Step
• DDeploy the model by using an A/B testing strategy that uses Amazon SageMaker Pipelines.
• EDeploy the model by using an A/B testing strategy that uses Amazon SageMaker and AWS Step

• A
  17% (5)
• B
  66% (19)
• D
  10% (3)
• E
  7% (2)

Explanation

Amazon Forecast (B) includes built-in holidays featurization (automatically incorporating public holiday calendars by country/region) and a weather index feature that ingests weather data for geographic areas - exactly matching the requirement to predict demand by region factoring in holidays and weather, with no custom feature engineering needed. For deployment, a canary strategy (C) gradually shifts a small percentage of traffic to the new model over a controlled window (2–3 days in this case), automatically rolling back if metrics degrade. SageMaker and AWS Step Functions support this natively with minimal setup. The Prophet algorithm (A) is a general time-series model and lacks the built-in holiday/weather featurization of the Amazon Forecast service. A/B testing strategies (D, E) split traffic between two models simultaneously for comparison - they are not designed for the gradual, time-boxed rollout described, and carry more operational complexity than a canary deployment.

No community discussion yet for this question.