MLS-C01 Mock Exam 76 — 25 Questions

Question 1

A data scientist is performing EDA on a dataset with 50 features and 10,000 observations. They suspect multicollinearity among some features. Which method is most appropriate to detect and quantify multicollinearity?

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

Accepted Answer

Variance Inflation Factor (VIF). VIF is the most appropriate method because it provides feature-specific quantification of multicollinearity, detecting both pairwise and multivariate collinearity. It is widely used and interpretable, with clear thresholds. Other methods either only detect pairwise relationships or do not directly quantify collinearity for individual features.

Answer

Condition number

Answer

Correlation matrix

Answer

Principal Component Analysis (PCA)

Question 2

A machine learning engineer is developing a model to predict whether a customer will churn within the next month. The dataset contains 10,000 records with 50 features, and the target variable is binary. The engineer wants to quickly establish a baseline and compare multiple algorithms with minimal coding effort. Which Amazon SageMaker built-in algorithm should the engineer use for this task?

Accepted Answer

Linear Learner. For a binary classification problem like churn prediction with tabular data, Linear Learner is a built-in SageMaker algorithm that provides a quick baseline. It supports supervised learning and can handle the dataset size and feature count without extensive preprocessing. Other options are unsupervised or dimensionality reduction techniques, which do not directly predict a binary target.

Answer

Random Cut Forest

Answer

Principal Component Analysis (PCA)

Answer

K-Means

Question 3

A company stores raw training data in an Amazon S3 bucket that uses SSE-KMS with a customer managed key. An ML engineer's SageMaker training job fails immediately with an access denied error when it tries to read the input channel, even though the engineer's IAM user has s3:GetObject permission on the bucket. The engineer confirms the bucket policy allows the role. What is the most likely cause?

Accepted Answer

The SageMaker execution role lacks kms:Decrypt permission on the customer managed key used to encrypt the objects.. Reading SSE-KMS encrypted objects requires two layers of authorization: S3 object access and permission to use the KMS key. A SageMaker training job operates under its execution role, so the engineer's own s3:GetObject permission does not apply. Granting kms:Decrypt on the customer managed key to the execution role, and ensuring the key policy allows it, resolves the immediate access denied error.

Answer

The training job must be configured with an S3 VPC endpoint because the bucket is in a different Availability Zone.

Answer

SSE-KMS encrypted objects cannot be read by SageMaker training jobs and must be decrypted to SSE-S3 before use.

Answer

The S3 bucket policy must explicitly list the SageMaker service principal as a principal and allow s3:GetObject.

Question 4

A data scientist is using Amazon SageMaker Automatic Model Tuning to optimize a built-in XGBoost binary classification model. The tuning job has completed, and the scientist wants to identify the best training job. The tuning job objective metric is 'validation:auc'. Which two statements are correct regarding how SageMaker determines the best training job and how the scientist can retrieve its details? (Choose two.)

Accepted Answer

The best training job is the one that achieved the highest value of the objective metric across all completed training jobs, provided the metric is maximized.. SageMaker Automatic Model Tuning selects the best training job by comparing the objective metric across all completed jobs according to the specified goal (Maximize or Minimize). For 'validation:auc', higher is better. The DescribeHyperParameterTuningJob API returns a BestTrainingJob field that includes the name of the best job. The scientist can then use DescribeTrainingJob to retrieve its hyperparameters and model artifacts. The best job is not necessarily the last one, and manual comparison is not required.

Answer

The best training job is always the last training job launched by the tuning job, regardless of its objective metric value.

Answer

The scientist must manually compare the objective metric values of all training jobs in the AWS Management Console to determine the best one.

Answer

The best training job is determined by averaging the objective metric across all epochs, not by the final epoch value.

Question 5

A data scientist is performing exploratory data analysis on a SageMaker Studio notebook with a housing dataset. Several continuous features are heavily right-skewed, the target price column has a long tail, and a subset of rows contains missing values in multiple columns. The scientist wants to assess relationships between features and the target while preparing the data for a linear model. Which two actions are most appropriate? (Choose two.)

Accepted Answer

Visualize pairwise relationships with a scatterplot matrix and inspect the correlation matrix to detect nonlinearity and multicollinearity.. Transforming skewed features and the target makes correlation and linear coefficients reflect genuine relationships rather than being driven by extreme values, and visual plus correlation-based inspection exposes nonlinearity and redundancy that summary numbers conceal. Dropping incomplete rows, standardizing before correlation, and silent mean imputation either discard information, change nothing, or distort the statistics being examined.

Answer

Drop every row containing any missing value before computing any statistics.

Answer

Standardize all features to zero mean and unit variance before computing the correlation matrix.

Answer

Replace all missing values with the column mean and proceed without recording which values were imputed.

Question 6

A data scientist is using Amazon SageMaker to train a model with the built-in K-Means algorithm for customer segmentation. The dataset contains 50,000 records with 10 numerical features. The data scientist wants to determine the optimal number of clusters (k) and evaluate the clustering quality. Which approach should they use?

Accepted Answer

Train multiple K-Means models with different values of k and use the elbow method by plotting the 'test:msd' against k, then select the k at the elbow point.. To determine the optimal number of clusters in K-Means, the elbow method is commonly used. It involves training models with varying k and plotting the mean squared distance to centroid (test:msd) against k. The point where the rate of decrease sharply changes—the elbow—suggests a good balance between cluster compactness and number of clusters. SageMaker's K-Means algorithm outputs this metric, making it straightforward to implement.

Answer

Train a single K-Means model with a large k (e.g., 100) and then merge clusters based on their centroid distances until the desired number is reached.

Answer

Train multiple K-Means models with different values of k and compare the 'test:msd' metric (mean squared distance to centroid) from the training logs, selecting the k with the lowest value.

Answer

Use the SageMaker Automatic Model Tuning feature to tune k, with the objective metric set to 'test:msd' and the goal to minimize it.

Question 7

A company stores sensitive customer data in Amazon S3. A data engineer needs to run a one-time SQL query to analyze this data using Amazon Athena. The data must be encrypted at rest with a customer-managed AWS KMS key, and the engineer must ensure that Athena can decrypt the data. What should the engineer do to allow Athena to access the encrypted data?

Accepted Answer

Grant the IAM role used by Athena the kms:Decrypt permission on the KMS key and ensure the S3 bucket policy allows Athena access.. To query SSE-KMS encrypted data in S3 with Athena, the IAM role used by Athena must have kms:Decrypt permission on the customer-managed KMS key. The S3 bucket policy must also allow Athena to read the objects. This ensures that Athena can access the encrypted data and decrypt it using the KMS key, satisfying both security and access requirements.

Answer

Configure the S3 bucket to use SSE-S3 encryption instead of SSE-KMS, and grant Athena full S3 access.

Answer

Attach an S3 bucket policy that allows Athena to decrypt objects and grant the Athena IAM role s3:GetObject permission.

Answer

Use Amazon S3 Select to query the data directly, which bypasses the need for KMS permissions.

Question 8

A data analyst wants to interactively explore a 40 GB dataset in Amazon S3 with SQL and immediately see result tables and simple charts, without provisioning a database or writing ETL code. The analyst has no preference for a specific query engine and wants the least operational effort. Which AWS service should the analyst use?

Accepted Answer

Amazon Athena with a workgroup configured to output query results to S3.. Athena queries data in place in S3 using the Glue Data Catalog for schema, so no cluster or warehouse provisioning is needed. The analyst can run SQL, view paginated result sets in the console, and download or visualize results with a connected BI tool. EMR, Redshift Spectrum, and RDS all require standing up and managing compute or storage resources that exceed the stated requirement.

Answer

Amazon RDS for PostgreSQL with the aws_s3 extension to import the S3 objects into tables.

Answer

Amazon EMR with a transient cluster running Apache Hive on the dataset in S3.

Answer

Amazon Redshift Spectrum with an external schema pointing at the S3 prefix.

Question 9

A data scientist is running an Amazon SageMaker Studio notebook to explore a 40 GB Parquet dataset stored in Amazon S3. Queries against the full dataset take several minutes and repeatedly exhaust the notebook instance memory. The scientist wants to interactively explore subsets and compute summary statistics without provisioning a larger instance or copying the data. Which approach best meets these requirements?

Accepted Answer

Use Amazon Athena to query the Parquet data in place and read the result sets from the notebook with the Athena API or JDBC driver.. Athena is the right fit because it separates compute from storage and queries Parquet in S3 in place, so the notebook never has to hold the dataset in memory. Columnar projection and predicate pushdown keep scans small, and results stream back in pages. The other approaches either require moving or duplicating the data, or they still execute on a memory-bound notebook or ML instance.

Answer

Convert the Parquet files to CSV in S3 and load them into the notebook with pandas read_csv in chunks.

Answer

Enable Amazon SageMaker Data Wrangler and import the full dataset into a Data Wrangler flow for interactive analysis.

Answer

Register the S3 prefix as an Amazon SageMaker Feature Store feature group and query it with the SageMaker Studio feature store UI.

Question 10

A machine learning engineer is training a SageMaker built-in Object Detection model on a dataset of 10,000 images with 20 object classes. The training job uses a p3.8xlarge instance and completes successfully, but validation mAP is only 0.15. The engineer suspects the default anchor box configuration is poorly matched to the dataset. Which SageMaker Object Detection hyperparameter should be adjusted to improve anchor box alignment with the ground truth bounding boxes?

Accepted Answer

aspect_ratios. In the SageMaker built-in Object Detection algorithm, anchor boxes are defined by scales and aspect ratios. When the default aspect ratios do not match the shapes of objects in the dataset, the anchors have low overlap with ground truth boxes, causing poor localization and low mAP. Adjusting aspect_ratios to reflect the actual object dimensions improves anchor alignment. Other hyperparameters like num_classes, mini_batch_size, and learning_rate do not change anchor geometry.

Answer

learning_rate

Answer

num_classes

Answer

mini_batch_size

Question 11

A machine learning engineer is training a linear regression model with Amazon SageMaker's built-in Linear Learner algorithm on a dataset where the target variable has a skewed distribution and contains a few extreme outliers. The engineer wants the model to be robust to these outliers without manually transforming the target. Which setting should the engineer configure?

Accepted Answer

Set predictor_type to 'regressor' and use the 'huber' loss function.. The huber loss is a robust regression loss that behaves like squared error for small errors and like absolute error for large errors, limiting the influence of outliers. Configuring the built-in Linear Learner with predictor_type='regressor' and loss='huber' directly addresses the skewed target and extreme values without manual transformation, aligning with the requirement.

Answer

Set predictor_type to 'regressor' and use the 'logistic' loss function.

Answer

Set predictor_type to 'regressor' and use the 'squared' loss function with L1 regularization.

Answer

Set predictor_type to 'regressor' and enable 'normalize_label' to True.

Question 12

A machine learning engineer is using Amazon SageMaker to train a deep learning model for a regression task. The model's training loss decreases steadily, but the validation loss starts to increase after a few epochs. The engineer wants to automatically stop training when validation loss stops improving to avoid overfitting. Which SageMaker feature should be used?

Accepted Answer

Enable early stopping in the training job configuration.. SageMaker early stopping is designed to halt training when a monitored metric, such as validation loss, fails to improve over a specified number of epochs. This prevents overfitting and reduces cost. The other options either require manual intervention, are not automated stopping mechanisms, or address regularization rather than stopping criteria.

Answer

Increase the dropout rate in the model architecture.

Answer

Set a lower learning rate to slow down training.

Answer

Use SageMaker Automatic Model Tuning (AMT) with a hyperparameter range for the number of epochs.

Question 13

A machine learning engineer is deploying a real-time inference service on Amazon SageMaker. The model must serve predictions with minimal latency, and the engineer needs to monitor the endpoint for data quality drift and model quality drift. The engineer also wants to receive automated alerts when drift exceeds a threshold. Which two actions should the engineer take to meet these requirements? (Choose two.)

Accepted Answer

Enable SageMaker Model Monitor on the endpoint with a data quality baseline and a monitoring schedule.. To monitor an endpoint for data and model quality drift, SageMaker Model Monitor is the appropriate service. It requires a baseline and a monitoring schedule to continuously evaluate data captured from the endpoint. The metrics emitted to CloudWatch can be used to create alarms that trigger SNS notifications, providing automated alerts. Together, these two actions satisfy the monitoring and alerting requirements.

Answer

Enable SageMaker Clarify to detect bias in the endpoint's predictions and send alerts via CloudWatch Events.

Answer

Configure the endpoint to use a multi-model server to reduce latency by loading only the required model.

Answer

Use SageMaker Debugger to capture tensors during inference and set built-in rules for drift detection.

Question 14

A company runs a real-time SageMaker endpoint backed by an Auto Scaling policy. Traffic is highly variable: it is near zero overnight and spikes sharply every weekday at 09:00. Users report timeouts during the first few minutes of each spike because the endpoint scales out too slowly. The engineer must reduce cold-start scale-out latency while keeping costs low during idle periods. Which change is MOST effective?

Accepted Answer

Configure a scheduled scaling policy that raises the desired instance count shortly before 09:00 each weekday, combined with the existing target tracking policy.. The spike is predictable, so proactive capacity is the right tool: a scheduled scaling policy in Application Auto Scaling pre-provisions instances before 09:00, while target tracking continues to handle unpredictable variation within the spike. Because the schedule also scales back down overnight, idle costs stay controlled, giving the best latency improvement for this recurring pattern.

Answer

Enable Elastic Inference on the endpoint and attach an accelerator to each instance to absorb the spike.

Answer

Increase the target value of the TargetTrackingScaling policy so the endpoint scales out more aggressively at lower utilization.

Answer

Switch the endpoint to Serverless Inference with a provisioned concurrency of zero and rely on automatic scaling.

Question 15

A machine learning engineer is using Amazon SageMaker to train a model on a dataset that contains sensitive financial information. The company requires that the data be encrypted at rest and in transit, and that the training job uses a specific AWS KMS customer managed key for encryption. The engineer has configured the training job to use the KMS key for the output volume. What additional step must the engineer take to ensure that the training data in Amazon S3 is encrypted with the same KMS key?

Accepted Answer

Ensure that the S3 bucket objects are encrypted with the desired KMS key by specifying it during upload or by setting the bucket's default encryption to that key.. The encryption of S3 objects is determined at upload or by bucket default encryption. To use a specific KMS key, the data must be uploaded with that key or the bucket must default to it. The SageMaker execution role then needs kms:Decrypt permission for that key. Specifying the key in the training job input or using SSE-S3 does not achieve the requirement.

Answer

Specify the KMS key in the SageMaker training job's input data configuration using the S3DataSource parameter.

Answer

Attach an IAM policy to the SageMaker execution role that allows s3:GetObject with the condition that the object is encrypted with the KMS key.

Answer

Enable default encryption on the S3 bucket using SSE-S3 (AES-256) instead of a KMS key.

Question 16

A machine learning team stores training data as approximately 4 million small JSON objects in an Amazon S3 bucket. A SageMaker training job that reads this prefix with File mode takes many hours just to start because the training instances must download every object before training begins. The team wants to reduce startup time without changing the training algorithm or the object contents. Which solution should a machine learning engineer recommend?

Accepted Answer

Convert the dataset to fewer, larger files such as sharded Parquet or RecordIO and configure the training job to use Pipe mode or Fast File mode.. Training performance on Amazon S3 data is limited by per-object overhead, not just raw bandwidth. Millions of tiny JSON files force the training container to issue millions of GET requests before training can start. Aggregating the records into a few large shards and streaming them with Pipe mode or Fast File mode removes both the request overhead and the full local download step, so the job starts far sooner.

Answer

Add an S3 Lifecycle rule that transitions the JSON objects to S3 Glacier Instant Retrieval, then rerun the training job.

Answer

Enable S3 Transfer Acceleration on the bucket and rerun the training job from the same Region.

Answer

Increase the number of training instances and enable network isolation on the training job.

Question 17

A company runs a nightly Apache Spark job on Amazon EMR that reads 5 TB of JSON from Amazon S3, joins it with a 2 GB reference table, and writes Parquet. The job has run reliably for months, but after the reference table grew, executors now fail with out-of-memory errors during the join stage. The engineer must reduce memory pressure without changing the output. Which approach is MOST appropriate?

Accepted Answer

Broadcast the reference table to all executors so the join avoids a shuffle.. When one join side is small enough to fit in executor memory, broadcasting it converts a shuffle join into a map-side join, removing the need to materialize the large side per partition. This directly relieves the out-of-memory condition caused by the grown reference table, as long as the broadcast fits within Spark's configured threshold.

Answer

Convert the reference table from Parquet to CSV to reduce its in-memory footprint.

Answer

Repartition the large JSON dataset by the join key before the join to increase parallelism.

Answer

Increase the number of executors and reduce each executor's memory allocation proportionally.

Question 18

A retail analytics team wants a fully managed, serverless way to run Spark ETL that reads from the AWS Glue Data Catalog and writes transformed Parquet back to Amazon S3 on a nightly schedule. They want to minimize operational overhead and pay only for the time the job runs. Which AWS service should they use?

Accepted Answer

AWS Glue jobs with a Spark ETL script, scheduled with a Glue trigger and using the Data Catalog for source and target tables.. AWS Glue provides a serverless Apache Spark environment that integrates directly with the AWS Glue Data Catalog for source and target schemas, and Glue triggers can run jobs on a nightly schedule. Because billing is based on DPU-hours consumed during job execution, the team avoids paying for idle capacity while keeping operational overhead low.

Answer

Amazon Athena with CREATE TABLE AS SELECT statements scheduled by Amazon EventBridge.

Answer

AWS Step Functions calling an AWS Lambda function that uses pandas for the transformation.

Answer

Amazon EMR running on Amazon EC2 with a persistent cluster and a nightly cron job.

Question 19

A financial institution is training a fraud detection model using Amazon SageMaker. The dataset is extremely imbalanced, with only 0.1% of transactions being fraudulent. The data scientist wants to maximize the model's ability to identify fraudulent transactions while minimizing false positives. Which evaluation metric should the data scientist use to optimize the model's performance?

Accepted Answer

Area Under the Precision-Recall Curve (AUC-PR). For highly imbalanced datasets, AUC-PR is preferred because it focuses on the minority class and the trade-off between precision and recall. AUC-ROC can be misleading due to the large number of true negatives. Accuracy is not suitable. While F1 score is useful, it evaluates a single threshold and does not capture the overall performance across thresholds as AUC-PR does.

Answer

Accuracy

Answer

F1 score with a threshold of 0.5

Answer

Area Under the ROC Curve (AUC-ROC)

Question 20

A machine learning engineer is using Amazon SageMaker to train a model. The training job is taking longer than expected, and the engineer suspects that the data loading is the bottleneck. The training data is stored in Amazon S3 as many small files. Which of the following is the most efficient way to improve data loading performance?

Accepted Answer

Use SageMaker Pipe mode to stream data directly from S3.. SageMaker Pipe mode streams data directly from S3 to the training container, eliminating the need to download all data before training begins. This is particularly beneficial for many small files, as it reduces startup latency and improves I/O throughput. Other options either do not address the bottleneck or introduce additional overhead.

Answer

Use a larger instance type with more CPU and memory.

Answer

Increase the number of epochs to allow more time for data loading.

Answer

Download the data to the instance's Amazon EBS volume before training.