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

You are building a fraud detection model for an e-commerce platform. One of the features is 'purchase_amount', which ranges from $1 to $10,000. The data has a skewed distribution with many small purchases and a few very large ones. You need to normalize this feature for your model, which uses gradient descent. Which normalization technique(s) would be most suitable in Snowflake, considering the data characteristics and the need to handle potential future outliers?

• A. Min-Max scaling using the following SQL:
  
• B. Robust scaling using interquartile range (IQR) in a stored procedure with Python:
  
• C. Power Transformer (e.g., Yeo-Johnson) implemented with Snowpark Python:
  
• D. Z-score standardization using the following SQL:
  
• E. Unit Vector normalization (L2 Normalization) using SQL:
  

Which of the following statements are TRUE regarding the 'Data Understanding' and 'Data Preparation' steps within the Machine Learning lifecycle, specifically concerning handling data directly within Snowflake for a large, complex dataset?

• A. Data Preparation in Snowflake can involve feature engineering using SQL functions, creating aggregated features with window functions, and handling missing values using 'NVL' or 'COALESCE. Furthermore, Snowpark Python provides richer data manipulation using DataFrame APIs directly on Snowflake data.
• B. The 'Data Understanding' step is unnecessary when working with data stored in Snowflake because Snowflake automatically validates and cleans the data during ingestion.
• C. During Data Preparation, you should always prioritize creating a single, wide table containing all possible features to simplify the modeling process.
• D. Data Understanding primarily involves identifying potential data quality issues like missing values, outliers, and inconsistencies, and Snowflake features like 'QUALIFY and 'APPROX TOP can aid in this process.
• E. Data Preparation should always be performed outside of Snowflake using external tools to avoid impacting Snowflake performance.

You are evaluating a binary classification model built in Snowflake for predicting customer churn. You have access to the model's predictions on a holdout dataset, and you want to use both the ROC curve and the confusion matrix to comprehensively assess its performance. Which of the following statements regarding the interpretation and use of ROC curves and confusion matrices are correct in this scenario?

• A. The ROC curve visualizes the trade-off between true positive rate (sensitivity) and false negative rate (1 - specificity) at various threshold settings.
• B. The area under the ROC curve (AUC) provides a single scalar value representing the overall discriminatory power of the model, with a higher AUC indicating better performance. An AUC of 0.5 indicates that the model performs no better than random chance.
• C. While the ROC curve is independent of the class distribution, the metrics derived from the confusion matrix (e.g., precision, recall) can be significantly affected by imbalanced datasets.
• D. In Snowflake, you can generate ROC curves and confusion matrices directly using the 'SYSTEM$PREDICT function with appropriate parameters and visualizing the results using a tool like Snowsight or Tableau.
• E. The confusion matrix allows you to calculate precision, recall, F I-score, and accuracy, which are all useful for understanding the model's performance in terms of correctly and incorrectly classified instances.

You're building a regression model using Snowpark Python to predict house prices. After initial training, you observe that the model consistently overestimates the prices of high-value houses and underestimates the prices of low-value houses. Given the options below, which optimization metric, along with code snippet to calculate it using Snowpark, would be most effective in addressing this specific issue?

• A. R-squared - as it measures the proportion of variance explained, directly addressing how well the model fits the data across all price ranges.
  
• B. Root Mean Squared Error (RMSE) - as it gives more weight to larger errors, making it suitable for addressing the underestimation/overestimation problem.
  
• C. Mean Squared Error (MSE) - as it is less sensitive to outliers than RMSE.
  
• D. Adjusted R-squared - as it penalizes the addition of irrelevant features, improving the model's generalization ability.
  
• E. Mean Absolute Error MAE - as it is sensitive to outliers and will penalize large errors more heavily.
  

A data scientist uses bootstrapping to estimate the sampling distribution of a statistic calculated from a dataset stored in Snowflake. They observe that the bootstrap distribution is significantly different from the original data distribution. Which of the following statements best describes the possible reasons for this difference, considering both the theoretical underpinnings of bootstrapping and potential limitations?

• A. Bootstrapping always provides accurate estimates of sampling distributions, any significant difference indicates an error in the code implementation.
• B. The difference is unexpected; the bootstrap distribution should always closely resemble the original data distribution, regardless of the statistic being estimated.
• C. Bootstrapping is only appropriate for normally distributed data; if the original data is not normal, the bootstrap distribution will inevitably differ significantly.
• D. The statistic being estimated is inherently unstable and has a high variance, causing the bootstrap distribution to be wider and potentially different in shape compared to the original data distribution. This is a normal outcome when dealing with such statistics.
• E. The original sample may not be representative of the population, and the bootstrap procedure is simply amplifying the biases present in the original sample. Additionally, the statistic itself may be highly sensitive to outliers or specific data points, leading to a distorted bootstrap distribution.