MLA-C01 Data Preparation for Machine Learning Practice Test 7 — 15 Questions

Question 1

A data scientist is preparing a dataset for a regression model using Amazon SageMaker Data Wrangler. The dataset contains missing values in several numerical columns. The scientist wants to impute these missing values with the mean of each column. Which Data Wrangler transform should be used to achieve this?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Handle missing values with the 'Impute missing values' transform and select 'Mean' as the imputation strategy.. Data Wrangler's 'Handle missing values' transform is designed specifically for imputation. It supports multiple strategies, including mean, median, mode, and custom values. By selecting 'Mean', the transform computes the mean of each numerical column and replaces missing entries with that mean. This is a standard and efficient way to handle missing numerical data in Data Wrangler.

Answer

Use the 'Drop missing values' transform to remove rows with missing data, then compute the mean from the remaining data.

Answer

Use the 'Encode categorical' transform to convert missing values to a separate category.

Answer

Apply the 'Standard scaler' transform to normalize the data, which will automatically fill missing values with zero.

Question 2

A data engineer is preparing a large dataset for a SageMaker training job. The dataset contains sensitive personally identifiable information (PII) in several columns. The engineer must anonymize the PII while preserving the statistical properties of the data for model training. The dataset is stored in Amazon S3 as Parquet files and is too large to fit in memory on a single instance. Which approach should the engineer use?

Accepted Answer

Use AWS Glue with a PySpark script to read the Parquet files, apply a hashing or tokenization transformation to the PII columns, and write the anonymized data back to S3.. AWS Glue with PySpark provides a distributed, scalable environment for processing large Parquet datasets that cannot fit in memory. It supports transformations such as hashing or tokenization to anonymize PII while preserving statistical properties. This approach is cost-effective and integrates with S3, making it suitable for preparing data for SageMaker training.

Answer

Use Amazon Macie to detect and redact the PII columns in place, then use the redacted data for training.

Answer

Use Amazon Athena to run a CREATE TABLE AS SELECT query that applies the SHA256 function to the PII columns, and store the results in a new S3 location.

Answer

Use SageMaker Data Wrangler to load the data, apply a custom transformation to hash the PII columns, and export the transformed data to S3.

Question 3

A data engineer is preparing a 4 TB dataset of Parquet files stored in Amazon S3 for a SageMaker training job. The training instances are GPU-based and the engineer wants to maximize throughput while minimizing S3 request costs across many epochs. Which data preparation approach BEST meets these goals?

Accepted Answer

Enable SageMaker Pipe mode and pass the S3 prefix directly to the training job's input channel.. Pipe mode is the SageMaker-native mechanism designed for large datasets because it streams shards from S3 into the training container without persisting the full dataset locally, cutting both disk requirements and repeated GET calls. Supplying the S3 prefix directly as an input channel activates this streaming behavior. The alternatives either duplicate data, add conversion overhead, or introduce a third-party filesystem layer that worsens request economics.

Answer

Convert the Parquet files to TFRecord and upload them to a second S3 bucket before training.

Answer

Mount the S3 bucket to the instance with s3fs and read files through the mounted filesystem.

Answer

Download the dataset to the instance's EBS volume with the AWS CLI in an on-start lifecycle script, then read from local disk.

Question 4

A data engineer is preparing features for a SageMaker model using SageMaker Data Wrangler. The flow reads a large Parquet dataset from S3, and the engineer wants the transformation steps to be reproducible outside the Data Wrangler UI and integrated into a CI/CD pipeline. Which action should the engineer take?

Accepted Answer

Export the Data Wrangler flow to a Python script and register it as a step in a SageMaker Pipeline.. Exporting the Data Wrangler flow to a Python script and registering it as a SageMaker Pipeline step turns the visual transformations into code that can be versioned, reviewed, and executed automatically. This satisfies both reproducibility and CI/CD integration, because the pipeline can be triggered on a schedule or by events and the transformation logic lives in source control.

Answer

Schedule the Data Wrangler flow to run on a recurring basis from the SageMaker Studio UI.

Answer

Use the Data Wrangler visual interface to apply the transformations, then download the transformed data and commit it to the repository.

Answer

Export the Data Wrangler flow to a Jupyter notebook that uses the SageMaker Data Wrangler library, then run it as a SageMaker Processing job.

Question 5

A data scientist is building a feature transformation with Amazon SageMaker Data Wrangler. The source CSV contains a column with categorical values, and the scientist needs the exported SageMaker Pipeline to apply the identical encoding during both training and later batch inference so that category-to-index mappings stay consistent. Which approach meets this requirement?

Accepted Answer

Export the data flow to a SageMaker Pipeline that includes the built-in categorical encoding transform step, and let the pipeline persist the fitted encoder artifact.. A reusable encoder must be fitted on training data and persisted so that the exact same category-to-index mapping is applied later. Exporting the Data Wrangler flow into a SageMaker Pipeline with the categorical encoding step preserves the fitted encoder as a pipeline artifact, which both the training and batch transform executions can reference, avoiding the column mismatch that arises when encoding is recomputed independently.

Answer

Export the data flow to a Python script that calls pandas.get_dummies on the column, then run the script as a Processing step in the pipeline.

Answer

Use the Data Wrangler quick model analysis to identify the important categories, then hard-code the index mapping in the training container's entry point.

Answer

Export the data flow to a SageMaker Feature Store feature group and enable online store, letting the store encode categories at write time.

Question 6

A data scientist is using SageMaker Data Wrangler to prepare a dataset for a binary classification model. The dataset contains a mix of numerical and categorical features, and some columns have missing values. The scientist wants to ensure that the data preparation steps are reproducible and can be integrated into a SageMaker Pipelines workflow. Which TWO actions should the scientist take? (Choose two.)

Accepted Answer

Use the Data Wrangler 'Export to SageMaker Pipelines' feature to generate a pipeline definition that includes the data preparation steps.. Data Wrangler flows can be exported as a Python script or directly as a SageMaker Pipelines definition. Both approaches capture the transformation steps in a reproducible, versionable artifact that can be executed as a processing step within a pipeline. This ensures consistency and automation, which are key for production ML workflows.

Answer

Configure the Data Wrangler flow to automatically retrain the model when new data arrives.

Answer

Manually record the data preparation steps in a notebook and run them as a separate processing job outside the pipeline.

Answer

Use the Data Wrangler built-in join and concatenate transforms to combine multiple data sources.

Question 7

A data engineer is using AWS Glue Studio to create an ETL job that joins two data sources: a large dataset in Amazon S3 (10 TB) and a smaller reference table in Amazon Redshift (100 GB). The engineer wants to optimize the join performance and minimize data transfer costs. Which strategy should be used?

Accepted Answer

Use a sort-merge join by partitioning both datasets on the join key and then merging.. For joining a large dataset with a smaller one, a sort-merge join is optimal when the smaller dataset cannot be broadcast. Partitioning both datasets on the join key allows parallel processing and minimizes data movement. This approach scales well for the 10 TB dataset and avoids the memory constraints of a broadcast join. It also reduces data transfer costs by processing data in place where possible.

Answer

Use a broadcast join by loading the Redshift table into memory on each worker node.

Answer

Use a cross join to combine all rows and then filter, as it simplifies the ETL logic.

Answer

Use a nested loop join by iterating over the smaller Redshift table for each row in the S3 dataset.

Question 8

A machine learning engineer is designing a feature engineering workflow in Amazon SageMaker Data Wrangler for a customer-churn model. The source data includes a high-cardinality customer_id column, a categorical plan_type column with 12 distinct values, and a numeric monthly_spend column with a long right tail. The team wants the exported pipeline to be reproducible and to avoid target leakage. Which TWO preprocessing decisions are appropriate? (Choose two.)

Accepted Answer

Standardize monthly_spend with a z-score transform and handle the tail with a quantile or log transform.. The defensible choices are the ones that reduce dimensionality and distribution skew while keeping all learned statistics confined to training data. One-hot encoding a low-cardinality categorical feature and removing a non-generalizable identifier keep the feature set clean, and combining standardization with a tail-compressing transform stabilizes numeric inputs. The rejected options all introduce leakage or unnecessary duplication into an exported flow.

Answer

Impute missing monthly_spend values with the column mean computed after the train-test split on the full dataset.

Answer

Balance the churn label by duplicating minority-class rows inside the Data Wrangler flow before export.

Answer

Encode customer_id with a frequency or target encoding so the model can exploit repeat-customer patterns.

Question 9

A data engineer is preparing a feature set for a SageMaker model and needs to handle missing values, encode categorical variables, and split the data into train, validation, and test sets. The team wants the transformation logic to be reproducible and reusable across training and batch inference. Which TWO approaches satisfy these requirements? (Choose two.)

Accepted Answer

Author the transformations in SageMaker Data Wrangler and export the flow to a SageMaker Pipeline so the same steps run for training and inference.. Reproducibility requires the transformation logic to be captured as an artifact that is fit on training data and replayed unchanged elsewhere. Exporting a Data Wrangler flow into a SageMaker Pipeline and serializing a fitted scikit-learn Pipeline to Amazon S3 both produce versioned, reusable transformers that apply identical imputation, encoding, and split behavior to training and inference data.

Answer

Apply pandas fillna and get_dummies to the full dataset before splitting, then save each split as a separate CSV.

Answer

Create a SageMaker Processing job with a script that reads the data and writes splits, relying on SageMaker's automatic feature store integration to persist the transformations.

Answer

Write the transformations as SQL in Amazon Athena and materialize the train, validation, and test splits as separate tables.

Question 10

A data engineer is cataloging a dataset in the AWS Glue Data Catalog so that Amazon Athena and SageMaker can query it consistently. The data is stored in S3 as newline-delimited JSON with nested fields, and new partitions are added daily. The engineer wants Athena queries to work without redefining the schema each day. Which combination of steps is MOST appropriate?

Accepted Answer

Define the table manually in the Glue Data Catalog with the correct nested schema, and enable partition projection on the timestamp column.. Manually defining the table with an accurate nested schema prevents crawler mis-inference on JSON, and partition projection removes the need to register each daily partition in the catalog. Athena then resolves partition paths from the projection configuration at query time, so new data is immediately queryable and the schema stays consistent without recurring crawler runs.

Answer

Load the JSON into Amazon Redshift using COPY, and query Redshift from Athena through a federated connector.

Answer

Create an AWS Glue crawler that points at the S3 prefix, runs on a daily schedule, and writes to a database in the Glue Data Catalog.

Answer

Create an external table in Amazon RDS that references the S3 location, and query it from Athena.

Question 11

A data engineer must join a 4 GB customer reference table with a 900 GB clickstream table in AWS Glue before writing curated data to Amazon S3. The join key is evenly distributed, and the engineer wants to avoid shuffling the large table across the cluster. Which Glue ETL technique should the engineer use?

Accepted Answer

Use a broadcast join by broadcasting the 4 GB customer reference table and joining it against the large clickstream DataFrame.. When one side of a join is small enough to fit in executor memory, broadcasting it eliminates the shuffle of the large side. Here the 4 GB customer table is the small relation and the 900 GB clickstream table is the large one, so broadcasting the customer table and streaming the clickstream data through the join avoids repartitioning the massive dataset and is the appropriate Glue ETL design.

Answer

Configure the job with the bookmark option enabled and rely on incremental processing to reduce the join size.

Answer

Repartition both DataFrames on the join key with 200 partitions and let the shuffle sort merge join proceed.

Answer

Use a broadcast join by placing the large table in the broadcast variable and the small table in the main DataFrame.

Question 12

A data scientist is preparing a training dataset in Amazon SageMaker Data Wrangler and wants to detect data quality problems before exporting the flow to a SageMaker Pipeline. Which two Data Wrangler capabilities help identify issues such as target leakage and anomalous values during analysis? (Choose two.)

Accepted Answer

The Data Quality and Insights Report, which summarizes missing values, duplicate rows, and anomalous feature distributions.. Data Wrangler provides purpose-built analyses that run against the imported dataset before transformations are exported. The Data Quality and Insights Report surfaces missing values, duplicates, imbalance, and anomalous distributions, while target leakage analysis flags features that appear too predictive of the label. Together they let the scientist find and remediate quality and leakage issues during analysis rather than discovering them after the pipeline is in production.

Answer

The Quick Model transform, which trains a model on the full dataset and reports the accuracy that will be achieved in production.

Answer

The custom transform using PySpark, which runs arbitrary code that can compute any statistic the scientist writes.

Answer

The bias report, which measures disparate impact across sensitive groups and corrects the imbalance automatically.

Question 13

A data scientist is preparing a large dataset for training a machine learning model. The dataset contains missing values in several columns. Which approach is the MOST efficient for handling missing values in a large dataset using AWS services?

Accepted Answer

Use Amazon SageMaker Data Wrangler to impute missing values using built-in transforms.. Amazon SageMaker Data Wrangler provides a visual interface and built-in transforms for handling missing values efficiently at scale, without writing custom code. Glue ETL is more code-heavy, and imputation with pandas is not scalable for large datasets. Removing all rows with missing values is not always optimal and may not be efficient.

Answer

Use AWS Glue ETL to write a custom Python script that imputes missing values with the mean.

Answer

Use pandas in a SageMaker notebook to impute missing values with the median.

Answer

Remove all rows with missing values from the dataset.

Question 14

A data scientist is preparing a dataset for a regression model using Amazon SageMaker Data Wrangler. The dataset contains missing values in several numerical columns. The scientist wants to impute these missing values with the mean of each column. Which Data Wrangler transform should be used to achieve this?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Remove the column from the feature set.. 'CustomerID' is a unique identifier with no predictive power for churn. Including it as a feature would cause the model to memorize individual customers rather than learn generalizable patterns, leading to overfitting and poor performance on unseen data. In machine learning, such columns should be removed during data preparation to ensure the model learns from meaningful features.

Answer

Keep the column as a feature because it uniquely identifies each customer.

Answer

Use the column as the target variable.

Answer

Encode the column using one-hot encoding.

Question 15

A data scientist is using Amazon SageMaker Data Wrangler to prepare a dataset. The dataset contains a column with date strings in the format 'YYYY-MM-DD'. The data scientist wants to extract the year, month, and day as separate features. Which Data Wrangler transform should be used?

Accepted Answer

Parse date transform.. The 'Parse date' transform in Amazon SageMaker Data Wrangler is specifically designed to convert date strings into structured datetime components. By applying this transform to the 'YYYY-MM-DD' column, the data scientist can automatically extract year, month, and day as separate features, enabling downstream feature engineering without manual string parsing.

Answer

Encode categorical transform.

Answer

Scale values transform.

Answer

Handle missing transform.