Consider you are working on a credit risk scoring model using Snowflake. You have a table 'credit data' with the following schema: 'customer id', 'age', 'income', 'credit_score', 'loan_amount', 'loan_duration', 'defaulted'. You want to create several new features using Snowflake SQL to improve your model. Which combination of the following SQL statements will successfully create features for age groups, income-to-loan ratio, and interaction between credit score and loan amount using SQL in Snowflake? Choose all that apply.

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

A financial institution wants to predict fraudulent transactions on credit card data stored in Snowflake. The dataset includes features like transaction amount, merchant ID, location, time of day, and user profile information. The target variable is 'is_fraudulent' (0 or 1). You have trained several binary classification models (Logistic Regression, Random Forest, and Gradient Boosting) using scikit-learn and persisted them using a Snowflake external function for inference. To optimize for both performance (inference speed) and accuracy, which of the following steps should you consider before deploying your model for real-time scoring using the external function? SELECT ALL THAT APPLY.

• A. Evaluate the models on a representative held-out dataset within Snowflake using SQL queries (e.g., calculating AUC, precision, recall) to choose the model with the best balance of performance and accuracy before deploying it.
• B. Implement feature selection techniques (e.g., using feature importance scores from Random Forest or Gradient Boosting) to reduce the number of features passed to the external function, improving inference speed.
• C. Replace the trained models with a simple rule-based system based solely on transaction amount. If the amount is greater than a threshold, flag it as fraudulent, as this will be faster than calling the external function.
• D. Increase the batch size of requests sent to the external function to amortize the overhead of invoking the external function itself, even if it increases latency for individual transactions.
• E. Normalize or standardize the input features in Snowflake using SQL before passing them to the external function to ensure consistent scaling and potentially improve model performance.

You're building a linear regression model in Snowflake to predict house prices. You have the following features: 'square_footage', 'number of bedrooms', 'location id', and 'year built'. 'location id' is a categorical variable representing different neighborhoods. You suspect that the relationship between 'square footage' and 'price' might differ based on the 'location id'. Which of the following approaches in Snowflake are BEST suited to explore and model this potential interaction effect?

• A. Create interaction terms by multiplying 'square_footage' with one-hot encoded columns derived from 'location_id'. Include these interaction terms in the linear regression model.
• B. Fit separate linear regression models for each unique 'location_id', using 'square_footage', 'number_of_bedrooms', and 'year_built' as independent variables.
• C. Use the 'QUALIFY clause in Snowflake SQL to filter the data based on 'location_id' before calculating regression coefficients. This is incorrect approach.
• D. Create interaction terms by adding 'square_footage' and one-hot encoded columns derived from 'location_id'. Include these interaction terms in the linear regression model.
• E. Apply a power transformation to 'square_footage' before including it in the linear regression model. This correct, but only to one variable.

You've built a regression model in Snowflake to predict customer churn. You've calculated the R-squared score on your test data and found it to be 0.65. However, after deploying the model to production and monitoring its performance over several weeks, you notice the model's predictive accuracy has significantly decreased. Which of the following factors could contribute to this performance degradation?
Select all that apply.

• A. Overfitting: The model learned the training data too well, capturing noise and specific patterns that do not generalize to new data.
• B. Feature engineering inconsistencies: The feature engineering steps applied to the production data are different from those applied during training.
• C. Bias Variance trade off : Model is having high bias.
• D. Data drift: The distribution of the input features in the production data has changed significantly compared to the training data.
• E. Increased data volume: The production data volume has increased significantly, causing resource contention and impacting model performance in Snowflake.

You are tasked with deploying a real-time fraud detection model in Snowflake. The model requires very low latency (under 100ms) to prevent fraudulent transactions. The input data is streamed into a Snowflake table. You are considering using either a Scalar or Vectorized Python UDF for scoring. Which of the following approaches and considerations are MOST critical for achieving the desired performance and reliability? Assume the model itself is computationally inexpensive. Select all that apply.

• A. Use a Scalar UDF because it has lower overhead per invocation compared to a Vectorized UDF when processing individual transactions.
• B. Configure Snowflake's Auto-Suspend feature to aggressively suspend the warehouse when idle, to minimize costs.
• C. Pre-load the model into a static variable within the UDF code, ensuring it's only loaded once per worker node.
• D. Utilize Snowflake's Materialized Views to pre-compute frequently used features, reducing the amount of data the UDF needs to process.
• E. Use a Vectorized UDF with a small 'MAX BATCH_SIZE to minimize latency while still leveraging vectorization benefits.