DAA-C01 Question 104
Single answerML functionsA retail analytics team stores daily product sales in Snowflake and wants to create a next-7-day demand forecast for each product-store combination directly in SQL. They need a solution that minimizes data movement, supports forecasting from time-series data already in Snowflake, and can be operationalized by analysts without building a custom external ML pipeline. Which approach best fits this requirement?
- A
Use the built-in forecasting ML function in Snowflake to generate forecasts from the historical time-series data grouped by product and store.
- B
Export the sales data to an external notebook environment, train a custom ARIMA model there, and write the predictions back to Snowflake because Snowflake ML functions cannot perform forecasting.
- C
Use a classification ML function in Snowflake to predict the next 7 days of numeric demand because classification models are designed for future-value estimation.
- D
Create a materialized view over the sales table and rely on automatic refresh to produce future demand estimates without using any ML function.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Snowflake's native forecasting ML capability for time-series prediction directly on data stored in Snowflake. This is the most practical choice when the goal is to keep data in-platform, reduce operational complexity, and enable SQL-centric analysts to build forecasts without managing external training infrastructure. In Snowflake documentation and product guidance, built-in ML functions are intended for common predictive tasks such as forecasting and anomaly detection on data already in Snowflake. By contrast, exporting data to external tools adds unnecessary movement and orchestration overhead, classification is the wrong model category for numeric future demand prediction, and materialized views do not provide machine learning or forecasting behavior.
- A. Correct.
Correct. Snowflake provides built-in ML functions for common predictive tasks, including forecasting, that can be applied directly to data stored in Snowflake. This approach aligns with the requirement to minimize data movement and enable analysts to work in SQL. For a retail time-series use case with historical sales by product-store combination, the forecasting function is the most appropriate native option.
- B. Incorrect.
Incorrect. Exporting data to an external environment is possible in some architectures, but it does not best fit the stated requirement. The scenario specifically prioritizes minimizing data movement and avoiding a custom external ML pipeline. Also, the premise is wrong because Snowflake does provide native ML functionality for forecasting use cases.
- C. Incorrect.
Incorrect. Classification functions are intended to predict categorical outcomes, such as churn/no churn or fraud/not fraud. Forecasting numeric demand over future dates is a time-series regression/forecasting problem, not a classification problem. Choosing classification here reflects a misunderstanding of ML task types.
- D. Incorrect.
Incorrect. Materialized views improve query performance for repeated computations over existing data, but they do not generate predictive outputs for future periods. Automatic refresh keeps the view synchronized with underlying data; it does not perform forecasting or infer unseen future values.