COF-C03 Question 45
Single answerML modelsA retail company stores sales history in Snowflake and wants analysts to generate product demand forecasts without moving data to another platform. The team also wants to avoid provisioning separate infrastructure for model inference. Which Snowflake capability best meets this requirement?
- A
Create a Python UDF in Snowflake and train a custom forecasting model directly inside the UDF
- B
Use Snowflake Cortex AISQL functions to perform forecasting directly in Snowflake SQL
- C
Export the data to an external ML platform, train a model there, and load predictions back into Snowflake
- D
Use a materialized view to automatically generate time-series forecasts from historical sales data
Show answer and explanation
Correct answer: B
Explanation
The best answer is to use Snowflake Cortex AISQL functions for forecasting directly in Snowflake. For SnowPro Core, candidates should understand that Snowflake includes built-in AI capabilities that let users apply ML-style functions close to the data, reducing data movement and operational overhead. In this scenario, the business specifically wants analysts to generate forecasts in Snowflake and avoid managing separate serving infrastructure, which is exactly the kind of use case built-in Cortex functionality is designed to address. By contrast, exporting data to an external ML system adds complexity, and database features such as materialized views do not provide predictive modeling. Snowflake documentation on Cortex AI and SQL-accessible AI functions is the relevant reference area for this type of capability.
- A. Incorrect.
Incorrect. Python UDFs can execute Python logic, but they are not intended for training and managing custom ML forecasting workflows as a built-in, end-to-end Snowflake forecasting capability. This option also implies creating and operating custom model logic, which does not best satisfy the requirement to minimize infrastructure and keep forecasting simple for analysts.
- B. Correct.
Correct. Snowflake Cortex provides built-in AI/ML capabilities accessible from SQL, including forecasting use cases, allowing teams to generate predictions without exporting data or managing separate inference infrastructure. This aligns with the requirement for analysts to work directly in Snowflake and avoid external platforms.
- C. Incorrect.
Incorrect. Exporting data to an external ML platform is a common pattern, but it conflicts with the stated goal of avoiding data movement and separate infrastructure for inference. It adds operational complexity and weakens the core benefit of keeping analytics and ML close to the data in Snowflake.
- D. Incorrect.
Incorrect. Materialized views improve query performance for stored query results, but they do not create ML models or produce forecasts. This distractor reflects a common misconception that performance-oriented database objects can substitute for predictive analytics capabilities.