DSA-C03 試験問題を無料オンラインアクセス

You are building a data science pipeline in Snowflake to predict customer churn. The pipeline involves extracting data, transforming it using Dynamic Tables, training a model using Snowpark ML, and deploying the model for inference. The raw data arrives in a Snowflake stage daily as Parquet files. You want to optimize the pipeline for cost and performance. Which of the following strategies are MOST effective, considering resource utilization and potential data staleness?

• A.Implement a series of smaller Dynamic Tables, each responsible for a specific transformation step, with well-defined refresh intervals tailored to the data's volatility and the downstream model's requirements.
• B.Schedule all data transformations and model training as a single large Snowpark Python script executed by a Snowflake task, ignoring data freshness requirements.
• C.Load all data into traditional Snowflake tables and use scheduled tasks with stored procedures written in Python to perform the transformations and model training.
• D.Use a single, large Dynamic Table to perform all transformations in one step, relying on Snowflake's optimization to handle dependencies and incremental updates.
• E.Use a combination of Dynamic Tables for feature engineering and Snowpark ML for model training and deployment, ensuring proper dependency management and refresh intervals for each Dynamic Table based on data freshness requirements.

You have deployed a fraud detection model in Snowflake that predicts the probability of a transaction being fraudulent. After a month, you observe that the model's precision has significantly dropped. You suspect data drift. Which of the following actions would be MOST effective in identifying and quantifying the data drift in Snowflake, assuming you have access to the transaction data before and after deployment?

• A.Create a UDF in Snowflake to calculate the Kolmogorov-Smirnov (KS) statistic for each feature between the training data and the recent transaction data. Then, create an alert if the KS statistic exceeds a predefined threshold for any feature.
• B.Retrain the model daily with the most recent transaction data without performing any explicit data drift analysis, relying on the model to adapt to the changes.
• C.Periodically sample a small subset of the recent transaction data and manually compare it with the training data using descriptive statistics (mean, standard deviation).
• D.Use Snowflake's built-in profiling capabilities to generate summary statistics for the training data. Compare these summary statistics with the statistics generated for recent transaction data. If significant differences are observed, assume data drift.
• E.Calculate the Jensen-Shannon Divergence between the probability distributions of predicted fraud scores on the training set and the current production data set.

You're building a fraud detection model and want to determine if the average transaction amount for fraudulent transactions is significantly higher than the average transaction amount for legitimate transactions. You have two tables in Snowflake:
'FRAUDULENT TRANSACTIONS and 'LEGITIMATE TRANSACTIONS, both with a 'TRANSACTION AMOUNT column. You believe that FRAUDULENT TRANSACTIONS contains fewer than 30 transactions. You don't know the population standard deviations. What are the proper steps to conduct the hypothesis test, and what is the correct hypothesis statement?

• A.Perform a t-test. Null Hypothesis: The average transaction amount for fraudulent transactions is less than or equal to the average transaction amount for legitimate transactions. Alternative Hypothesis: The average transaction amount for fraudulent transactions is greater than the average transaction amount for legitimate transactions.
• B.Perform a chi-squared test. Null Hypothesis: There is no relationship between transaction amount and whether a transaction is fraudulent. Alternative Hypothesis: There is a relationship between transaction amount and whether a transaction is fraudulent.
• C.Perform a Z-test. Null Hypothesis: The average transaction amount for fraudulent transactions is equal to the average transaction amount for legitimate transactions. Alternative Hypothesis: The average transaction amount for fraudulent transactions is not equal to the average transaction amount for legitimate transactions.
• D.Perform a Z-test. Null Hypothesis: The average transaction amount for fraudulent transactions is less than or equal to the average transaction amount for legitimate transactions. Alternative Hypothesis: The average transaction amount for fraudulent transactions is greater than the average transaction amount for legitimate transactions.
• E.Perform a t-test. Null Hypothesis: The average transaction amount for fraudulent transactions is equal to the average transaction amount for legitimate transactions. Alternative Hypothesis: The average transaction amount for fraudulent transactions is not equal to the average transaction amount for legitimate transactions.

A marketing team is using Snowflake to store customer data including demographics, purchase history, and website activity. They want to perform customer segmentation using hierarchical clustering. Considering performance and scalability with very large datasets, which of the following strategies is the MOST suitable approach?

• A.Directly apply an agglomerative hierarchical clustering algorithm with complete linkage to the entire dataset within Snowflake, using SQL. This is computationally feasible due to SQL's efficiency.
• B.Employ BIRCH clustering with Snowflake Python UDF. Configure Snowflake resources accordingly. Optimize the clustering process. And tune parameters.
• C.Perform mini-batch K-means clustering using Snowflake's compute resources through a Snowpark DataFrame. Take a large sample of each mini-batch and perform hierarchical clustering on each mini-batch and then create clusters of clusters.
• D.Utilize a SQL-based affinity propagation method directly within Snowflake. This removes the need for feature scaling and specialized hardware.
• E.Randomly sample a small subset of the customer data and perform hierarchical clustering on this subset using an external tool like R or Python with scikit-learn. Assume that results generalize well to the entire dataset. Avoid using Snowflake for this purpose.

You've deployed a fraud detection model in Snowflake using Snowpark. You are monitoring its performance and notice a significant decrease in recall, while precision remains high. This means the model is missing many fraudulent transactions. The training data was initially balanced, but you suspect that recent changes in user behavior have skewed the distribution of fraudulent vs. non-fraudulent transactions in production. Which of the following actions are MOST appropriate to address this issue and improve the model's performance, considering best practices for model retraining within the Snowflake ecosystem?

• A.Immediately shut down the model to prevent further inaccurate classifications. Investigate why the recall is low before any retraining is performed.
• B.Retrain the model using the original training data. Since the precision is high, the model's fundamental logic is still sound. A larger training dataset isn't necessary.
• C.Adjust the model's classification threshold to be more sensitive, even if it means accepting a slightly lower precision. This can be done directly within Snowflake using a SQL UDF that transforms the model's output probabilities.
• D.Implement a data drift monitoring system in Snowflake to automatically detect changes in the input features of the model. Trigger an automated retraining pipeline when significant drift is detected. This retraining should include recent production data with updated labels, but only if label data collection can be automated.
• E.Retrain the model using a dataset that includes recent production data, being sure to re-balance the dataset to maintain a roughly equal number of fraudulent and non-fraudulent transactions. Prioritize transactions from the last month.