Snowflake DSA-C03 Exam : SnowPro Advanced: Data Scientist Certification Exam

You have trained a complex machine learning model using Snowpark for Python and are now preparing it for production deployment using Snowpark Container Services. You have containerized the model and pushed it to a Snowflake-managed registry. However, you need to ensure that only authorized users can access and deploy this model. Which of the following actions MUST you take to secure your model in the Snowflake Model Registry, ensuring appropriate access control, and minimizing the risk of unauthorized deployment or modification?

• A. Grant the 'READ privilege on the container registry to all users who need to deploy the model. Create a custom role with the 'APPLY MASKING POLICY privilege and grant this role to the deployment team.
• B. Grant the 'USAGE privilege on the database and schema containing the model registry, grant the 'READ privilege on the registry itself, and grant the EXECUTE TASK' privilege to the deployment team for the deployment task.
• C. Create a custom role, grant the USAGE' privilege on the database and schema containing the model registry, grant the 'READ privilege on the registry, and then grant this custom role to only those users authorized to deploy the model. Consider masking sensitive model parameters using masking policies.
• D. Grant the 'USAGE privilege on the stage where the model files are stored to all users who need to deploy the model.
• E. Store the model outside of Snowflake managed registry and use external authentication to control access.

You have developed a customer churn prediction model using Python and deployed it as a Snowflake UDE You are monitoring its performance and notice a significant drop in accuracy over time. To address this, you need to implement automated model retraining with regular validation. Which of the following steps and validation techniques are MOST critical for ensuring the retrained model is effective and avoids overfitting to recent data? (Select THREE)

• A. Monitor the model's performance on a live dataset and trigger retraining only when the performance drops below a predefined threshold, using metrics like accuracy, precision, or recall. Save Model Performance to 'MODEL_PERFORMANCE.
• B. Use cross-validation techniques (e.g., k-fold cross-validation) during the retraining process to estimate the model's performance on unseen data and prevent overfitting. Evaluate on a held-out validation set.
• C. Update the UDF in place using 'CREATE OR REPLACE FUNCTION' immediately after retraining completes, regardless of the validation results.
• D. Retrain the model using the entire available dataset, as this will maximize the amount of data the model learns from.
• E. Implement a data drift detection mechanism. Monitor the distribution of input features over time and trigger retraining if significant drift is detected using tools such as Snowflake's Anomaly Detection features or custom drift metrics calculated in SQL.

Consider the following Snowflake SQL query used to calculate the RMSE for a regression model's predictions, where 'actual_value' is the actual value and 'predicted value' is the model's prediction. However, you notice that the RMSE calculation is incorrect due to an error in the query. Identify the error in the query and provide the corrected query. The table name is 'sales_predictions'.

Which of the following options represents the corrected query that accurately calculates the RMSE?

• A.
  
• B.
  
• C.
  
• D.
  
• E.
  

A financial institution wants to use Snowflake Cortex to analyze customer reviews and feedback extracted from various online sources to gauge customer sentiment towards their new mobile banking application. The goal is to identify positive, negative, and neutral sentiments, and also extract key phrases that drive these sentiments. Which of the following steps represent a viable workflow for achieving this using Snowflake Cortex and related functionalities?

• A. 1. Ingest the customer reviews into a Snowflake table. 2. Use the 'SNOWFLAKE.ML.PREDICT' function with a sentiment analysis model to determine the overall sentiment score for each review. 3. Apply a separate key phrase extraction model via 'SNOWFLAKE.ML.PREDICT' to identify important keywords in the reviews.
• B. 1. Ingest the customer reviews into a Snowflake table. 2. Create a custom JavaScript UDF that calls the Snowflake Cortex 'COMPLETE' endpoint with a prompt that asks for both sentiment and key phrases. 3. Store the results in a new Snowflake table.
• C. 1. Ingest the customer reviews into a Snowflake table. 2. Use Snowflake's built-in 'NLP_SENTIMENT' function (if available) or a similar UDF based on a pre- trained sentiment analysis model to get the sentiment score. 3. Use regular expressions in SQL to extract key phrases based on frequency and context.
• D. 1. Create a Streamlit application hosted externally that connects to the Snowflake database. 2. The Streamlit app uses a Python library like 'transformers' to perform sentiment analysis and key phrase extraction on the customer reviews read from Snowflake. 3. The results are then written back to a separate Snowflake table.
• E. 1. Ingest the customer reviews into a Snowflake table. 2. Use the 'SNOWFLAKML.PREDICT' function with the appropriate task-specific model to determine the sentiment score for each review. 3. Further fine-tune the sentiment model with customer review data to improve the score and accuracy.

A Data Scientist is designing a machine learning model to predict customer churn for a telecommunications company. They have access to various data sources, including call logs, billing information, customer demographics, and support tickets, all residing in separate Snowflake tables. The data scientist aims to minimize bias and ensure data quality during the data collection phase. Which of the following strategies would be MOST effective for collecting and preparing the data for model training?

• A. Directly use all available columns from each table without any preprocessing to avoid introducing bias.
• B. Randomly select a subset of data from each table to reduce computational complexity and speed up model training.
• C. Create a single, wide table by performing a series of INNER JOINs on all tables using customer ID as the primary key. Handle missing values by imputing with the mean for numerical columns and 'Unknown' for categorical columns.
• D. Perform exploratory data analysis (EDA) on each table to identify relevant features and potential biases. Use feature selection techniques to reduce dimensionality. Implement robust data validation checks to ensure data quality and consistency before joining the tables. Handle missing values strategically based on the specific column and its potential impact on the model.
• E. Use Snowflake's Data Marketplace to supplement the existing data with external datasets, regardless of their relevance to the churn prediction problem.