COF-C03 Question 135
Single answerSnowflake MLA retail analytics team stores customer transaction data in Snowflake and wants to build a churn prediction model without moving data to an external platform. The team also wants SQL users to be able to generate predictions from inside Snowflake once the model is trained. Which approach best meets these requirements?
- A
Use Snowflake ML to train the model in Snowflake and then invoke the model for inference directly from Snowflake.
- B
Export the data to a third-party notebook environment, train the model there, and store only the prediction results back in Snowflake.
- C
Use Snowpipe Streaming to automatically create and train the churn model as new data arrives.
- D
Create a materialized view on the transaction table; Snowflake will use it to generate churn predictions automatically.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Snowflake ML so the data science workflow stays close to the data and predictions can be generated inside Snowflake. For SnowPro Core, candidates should recognize the high-level value proposition of Snowflake ML: enabling machine learning workflows on data stored in Snowflake while reducing unnecessary data movement. Snowpipe Streaming is an ingestion feature, not an ML training service, and materialized views are query optimization objects rather than predictive modeling tools. A common best practice in Snowflake is to minimize data movement for governance, security, and operational simplicity.
- A. Correct.
Correct. Snowflake ML is designed to support machine learning workflows where data remains in Snowflake. A team can train supported models using Snowflake-managed compute and then perform inference from within Snowflake, which aligns with the requirement to avoid data movement and let SQL users consume predictions in the same platform.
- B. Incorrect.
Incorrect. Although exporting data to an external environment is a common pattern in many organizations, it does not meet the stated requirement to avoid moving data out of Snowflake. It also adds operational complexity, governance concerns, and duplication of data pipelines.
- C. Incorrect.
Incorrect. Snowpipe Streaming is used for low-latency data ingestion into Snowflake. It does not create or train machine learning models automatically. This option confuses data ingestion capabilities with Snowflake ML capabilities.
- D. Incorrect.
Incorrect. Materialized views improve query performance for repeated query patterns, but they do not train or execute machine learning models. This distractor reflects the misconception that performance optimization features can provide predictive analytics functionality.