COF-C03 Question 46
Single answerML modelsA retail analytics team stores sales history in Snowflake and wants business analysts to generate demand forecasts directly in SQL without exporting data to another platform. The team also wants to minimize data movement and operational overhead. Which Snowflake feature should they use to build and run the forecasting model?
- A
Create a forecast using Snowflake Cortex AISQL forecasting functions directly on data stored in Snowflake
- B
Train a model by creating a materialized view with historical sales data and querying the view for predicted values
- C
Use Time Travel to restore prior versions of the sales table and compare the restored rows to estimate future demand
- D
Create a secure share of the sales data and let consumers generate forecasts automatically from the shared tables
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Snowflake's native ML-oriented SQL capabilities for forecasting so analysts can generate predictions directly where the data already resides. For SnowPro Core, candidates should distinguish between core platform features that store, govern, or share data and specialized capabilities that perform ML tasks. Materialized views improve query performance, Time Travel supports historical access and recovery, and Secure Data Sharing distributes data securely, but none of those features perform forecasting. Snowflake documentation for Cortex AISQL and ML-related SQL workflows emphasizes minimizing data movement and enabling in-platform analytics and prediction use cases.
- A. Correct.
Correct. Snowflake provides Cortex AISQL capabilities for ML tasks such as forecasting directly in Snowflake. This approach aligns with the scenario requirements: business analysts can work in SQL, data remains in Snowflake, and operational overhead is reduced because there is no need to move data to an external ML platform. This is the most appropriate native option for demand forecasting in a Snowflake-centric workflow.
- B. Incorrect.
Incorrect. A materialized view can precompute and store query results for performance optimization, but it does not train or execute ML forecasting models. Someone might choose this option because materialized views can speed analytics workloads, but they are not a substitute for machine learning functionality.
- C. Incorrect.
Incorrect. Time Travel is used to access historical data versions, recover objects, or query past states of data. It is not an ML capability and cannot generate demand forecasts. This distractor targets the misconception that any historical-data feature can be used for predictive modeling.
- D. Incorrect.
Incorrect. Secure Data Sharing allows Snowflake objects and data to be shared across accounts without copying data, but it does not create or run forecasting models automatically. A shared dataset could be used by another party for modeling, but the sharing feature itself is not the forecasting solution.