Snowflake Certified SnowPro Specialty - Snowpark Sample Questions:

1. Given a Snowpark DataFrame 'df with a column named 'data' of VARIANT type, where the VARIANT contains JSON objects with nested fields. You need to extract the value of the nested field 'address.city' as a STRING and the value of as a DOUBLE, handling cases where either 'address' or 'items' might be missing. Which combination of Snowpark functions is best suited to achieve this robustly and efficiently?

A)

B)

C)

D)

E)

2. You have a complex data pipeline implemented using Snowpark Tasks in a Directed Acyclic Graph (DAG). One of the tasks, , depends on the successful completion of two parent tasks, and 'task B'. You need to implement error handling such that if 'task_R fails, 'task_C' should not be executed, but should still complete its execution regardless of status. If 'task B' fails, 'task_C' should not be executed. How do you configure the task dependencies and error handling in Snowflake to achieve this behavior?

A)

B)

C)

D)

E)

3. You have a Snowflake table 'PRODUCT REVIEWS' with columns 'product_id', 'review_text', 'review_date', and 'sentiment_score'. You are using Snowpark to build a data pipeline that analyzes these reviews and identifies potentially malicious or spam reviews. Based on a machine learning model deployed as a Snowflake UDF, you have identified a set of 'product_id' values that are associated with malicious reviews, and are stored in a python list called 'malicious_product_ids'. You need to delete all rows from the 'PRODUCT REVIEWS' table where the 'product_id' is present in the list. Assume 'malicious_product_ids' contains thousands of product IDs. Which of the following approaches is the MOST efficient and scalable for deleting these records using Snowpark?

A) Option E
B) Option D
C) Option B
D) Option C
E) Option A

4. A data engineering team is developing a Snowpark stored procedure in Python to perform anomaly detection on time-series data stored in a Snowflake table named 'sensor_readingS. The stored procedure needs to efficiently process large volumes of data and return only the rows identified as anomalies. Which of the following approaches would provide the most performant and scalable solution for operationalizing this stored procedure?

A) Create a UDF with a Scala implementation and use it inside the Snowpark stored procedure to detect anomalies using the Scala implementation for increased processing power.
B) Execute a SQL query from within the stored procedure using the Snowflake connector for Python to fetch the relevant data, then use a standard Python loop to iterate through the results and apply anomaly detection logic. Return the anomalous rows as a list of dictionaries.
C) Use the method to include a pre-trained anomaly detection model (pickled object) in the stored procedure's execution environment. Load the model, use it to predict on the data fetched using 'session.table(Y , and return a Snowpark DataFrame of anomalies.
D) Use the Snowpark API to directly perform anomaly detection calculations (e.g., rolling statistics, z-score calculations) on the 'sensor_readings' table within the stored procedure, leveraging Snowpark's distributed processing capabilities, and then return the resulting Snowpark DataFrame containing only the anomalies.
E) Load the entire 'sensor_readings' table into a Pandas DataFrame within the stored procedure, perform anomaly detection using a Python library like 'scikit-learn' , and then create a Snowpark DataFrame from the filtered Pandas DataFrame to return the results.

5. You are developing a Snowpark application to process large datasets. You want to leverage asynchronous jobs to improve performance and prevent blocking the main thread. You have the following code snippet:

A) Implement a callback function using 'job.on_success(callback_function)' and 'job.on_error(error_function)' to handle the result or any errors asynchronously.
B) Continuously check 'job.status' in a loop until it returns 'SUCCESS', then retrieve the result using 'job.result()'. This is less efficient due to polling.
C) Use to retrieve the result with a timeout of 30 seconds, catching 'TimeoutError' if the job takes too long. This is the only way to get the result safely.
D) Use 'job.getQueryld(Y to fetch the query ID and query the execution status directly from Snowflake using SQL. This is inefficient and bypasses the Snowpark API.
E) Call 'job.result(Y without any error handling. Snowflake will automatically handle any errors and return a null result.

Solutions: