MLS-C01 Practice Exam 58 — 20 Questions

Question 1

A media company uses Amazon SageMaker to train a deep learning model for video classification. The training job uses a single ml.p3.2xlarge instance and processes 50 GB of labeled video data stored in Amazon S3. The training completes successfully in 12 hours. However, the data scientists report that the model’s accuracy is lower than expected. They suspect the training data contains labeling errors. To improve model accuracy without incurring significant additional cost, they want to identify and remove mislabeled training examples before retraining. They have a small budget of $50 and need to complete the analysis within 2 hours. Which approach should the data scientists take?

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

Accepted Answer

Train a small, fast model on a random sample of the data (e.g., 1 GB) using a cheaper instance like ml.m5.xlarge, then use the model's prediction confidence to flag low-confidence examples as potential mislabels for manual review.. Training a small, fast model on a 1 GB random sample using a cheaper instance (ml.m5.xlarge) allows the team to quickly identify low-confidence predictions, which are strong indicators of mislabeled examples. This approach fits within the $50 budget and 2-hour time constraint, as it avoids processing the full 50 GB dataset and leverages a lightweight model for rapid iteration. By flagging only suspicious samples for manual review, the team can efficiently clean the training data without incurring the cost of re-labeling the entire dataset.

Answer

Use SageMaker Ground Truth to create a new labeling job for the entire dataset, then compare the new labels with the original labels to identify discrepancies.

Answer

Use SageMaker Clarify to generate a bias report for the training data and remove instances that contribute to bias.

Answer

Manually review all 50 GB of video data to correct labels.

Question 2

A data scientist needs to deploy a trained model to Amazon SageMaker for real-time inference. The model is stored as a .tar.gz file in Amazon S3. Which AWS service is used to create a SageMaker endpoint?

Accepted Answer

SageMaker Model and Endpoint Configuration. To deploy a model to a SageMaker real-time endpoint, you create a SageMaker Model resource (pointing to the model artifacts in S3 and the inference container image), then an Endpoint Configuration (specifying instance type, count, and variants), and finally an Endpoint that references that configuration. This three-step process is the native SageMaker deployment path for real-time inference.

Answer

AWS Lambda

Answer

AWS CloudFormation

Answer

Amazon ECS

Question 3

A data scientist is performing hyperparameter optimization for a gradient boosting model using Amazon SageMaker Automatic Model Tuning. The objective metric is 'validation:logloss'. Which TWO strategies can help the tuning job converge faster? (Choose TWO.)

Accepted Answer

Use Bayesian optimization strategy. Option A is correct because SageMaker Automatic Model Tuning's Bayesian optimization strategy builds a probabilistic model of the objective function and uses it to choose the next hyperparameter combination, so it typically finds good configurations in fewer trials than uninformed search, helping the job converge faster. Option E is correct because enabling early stopping (via the SageMaker early stopping mechanism, e.g., the 'EarlyStopping' training job parameter with a training script that emits the objective metric) automatically terminates poorly performing training jobs before they run all epochs, reducing wasted compute and allowing the tuning job to reach a good result sooner. Option B is not correct because increasing the number of tuning jobs (max_jobs) simply runs more trials and generally increases total time and cost rather than making convergence faster. Option C is not correct because increasing resource limits (e.g., larger instance types or more instances) speeds up individual training jobs but does not improve the tuning algorithm's convergence toward the optimal hyperparameters. Option D is not correct because random search is an uninformed strategy that typically requires more trials than Bayesian optimization to find comparable hyperparameters, so it does not help the tuning job converge faster.

Answer

Increase the number of tuning jobs

Answer

Increase the resource limits for each training job

Answer

Use random search strategy

Question 4

An AWS Glue job is failing with an error that it cannot access an S3 bucket. The IAM role attached to the Glue job is shown in the exhibit. What is the MOST likely cause of the failure?

Accepted Answer

The S3 bucket has a bucket policy that denies access to this role. Even if the IAM role has S3 permissions, an S3 bucket policy that explicitly denies access to that role will override any allow. AWS evaluates all policies (identity-based and resource-based) and a deny in any policy results in a final deny decision. The error indicates the Glue job cannot access the bucket, which is consistent with a bucket-level deny.

Answer

The role lacks S3 permissions

Answer

The role does not have permission to call S3 APIs

Answer

The trust policy does not allow Glue to assume the role

Question 5

A machine learning engineer is setting up a training job in Amazon SageMaker. Which THREE components are required to define a training job? (Choose three.)

Accepted Answer

Output data configuration (e.g., model artifact path).. The three required components are C, D, and E. Option C (Output data configuration, e.g., model artifact path) is required because SageMaker needs an S3 location to write the resulting model artifacts after training completes. Option D (An algorithm or custom container image) is required because the training job must know which algorithm or Docker image to run to execute the training code. Option E (Input data configuration, e.g., S3 path) is required because SageMaker must know where the training dataset resides (typically in S3) to feed it to the algorithm. Options A and B are not required: VPC configuration is optional and only used for network isolation, and hyperparameters are optional since many algorithms have defaults and some require none.

Answer

VPC configuration for network isolation.

Answer

Hyperparameters for the algorithm.

Question 6

A data scientist trains a model using Amazon SageMaker's built-in XGBoost algorithm. The model overfits on the training data. Which hyperparameter adjustment is MOST likely to reduce overfitting?

Accepted Answer

Increase the value of the lambda (L2 regularization) hyperparameter.. Increasing the lambda (L2 regularization) hyperparameter adds a penalty on the squared magnitude of the model weights, which discourages the model from fitting noise in the training data. This directly reduces overfitting by shrinking the influence of individual features, a standard regularization technique in XGBoost.

Answer

Increase the value of the max_depth hyperparameter.

Answer

Increase the value of the subsample hyperparameter to 1.0.

Answer

Increase the value of the num_round hyperparameter.

Question 7

A company wants to centralize logging from multiple AWS accounts and on-premises servers. The logs must be stored cost-effectively and be searchable. Which TWO services should be used? (Choose TWO.)

Accepted Answer

Amazon Athena. Amazon S3 (option E) is correct because it provides durable, low-cost object storage ideal for centralizing log data from multiple AWS accounts and on-premises servers, and it serves as the underlying data store that Athena queries in place. Amazon Athena (option C) is correct because it is a serverless query service that runs standard SQL directly against data in S3, making the centralized logs searchable without loading them into a database or managing infrastructure. Together, S3 plus Athena deliver the required cost-effective storage and searchability. Amazon CloudWatch Logs (A) is designed for log ingestion, monitoring, and metric filters rather than cheap long-term centralized storage and ad hoc SQL search across accounts. Amazon Redshift (B) is a data warehouse that requires loading data and is more expensive than querying S3 directly. Amazon Kinesis Data Streams (D) is for real-time streaming ingestion, not cost-effective storage or search.

Answer

Amazon CloudWatch Logs

Answer

Amazon Redshift

Answer

Amazon Kinesis Data Streams

Question 8

A company is using Amazon Redshift for data warehousing. The data engineering team observes that query performance degrades over time due to data skew. Which three strategies should the team implement to improve performance?

Accepted Answer

Choose appropriate distribution keys based on join and group-by columns.. Option A is correct because choosing a distribution key on columns frequently used in joins or GROUP BY clauses co-locates matching rows on the same slice, eliminating data movement (redistribution or broadcast) across nodes and directly mitigating skew-related slowdowns. Option C is correct because VACUUM reclaims space from deleted rows and re-sorts data to restore sort-key ordering, while ANALYZE refreshes table statistics so the query planner generates efficient plans; both are essential maintenance tasks as data changes over time. Option D is correct because defining appropriate sort keys (compound or interleaved) enables zone maps to skip irrelevant blocks during scans, reducing I/O and improving query performance on large tables. Option B is not correct because adding nodes increases capacity and parallelism but does not fix the underlying skew problem, and a poorly chosen distribution key can actually worsen skew across the additional slices. Option E is not correct because Amazon Redshift does not use traditional indexes; it relies on sort keys, distribution keys, and zone maps, so dropping indexes is not a valid Redshift tuning strategy.

Answer

Increase the number of nodes in the Redshift cluster.

Answer

Drop unused indexes on large tables.

Question 9

A company is building a classification model and discovers that the target variable is imbalanced: 95% of samples belong to class A and 5% to class B. The data scientist needs to understand the distribution of numeric features for each class. Which approach is most appropriate?

Accepted Answer

Use Amazon SageMaker Data Wrangler to create histograms for each feature, grouped by class label.. Amazon SageMaker Data Wrangler is designed for exploratory data analysis (EDA) and can quickly generate visualizations like histograms. By grouping histograms by the class label, the data scientist can directly compare the distribution of each numeric feature for class A versus class B, which is exactly what is needed to understand feature distributions in an imbalanced dataset. This approach is efficient, requires minimal code, and integrates seamlessly with SageMaker workflows.

Answer

Run a t-test for each feature to determine statistical significance between classes.

Answer

Generate box plots for each feature using Amazon QuickSight.

Answer

Compute the correlation matrix between features and the target.

Question 10

A data scientist is using an IAM role with the policy shown in the exhibit to train a model in SageMaker. The training job fails with a permissions error. What is the missing permission?

Accepted Answer

iam:PassRole. The training job fails because SageMaker needs to assume the IAM role specified in the training job configuration to access resources like S3 buckets. The `iam:PassRole` permission is required to allow the SageMaker service to pass that role to the training job. Without it, SageMaker cannot assume the role and thus cannot perform actions such as reading training data from S3.

Answer

sagemaker:InvokeEndpoint

Answer

sagemaker:DescribeTrainingJob

Answer

s3:ListBucket

Question 11

A data scientist is analyzing a dataset with 100 features and 10,000 samples. The target variable is highly imbalanced (1% positive class). Which exploratory data analysis step is most critical before model training?

Accepted Answer

Plot the histogram of the target variable. Plotting the histogram of the target variable is the most critical EDA step because it directly reveals the class imbalance (1% positive), which can severely impact model training and evaluation. Understanding the distribution helps in choosing appropriate techniques like resampling, class weights, or specialized metrics. Without this step, the data scientist might overlook the imbalance and build a model that predicts the majority class with high accuracy but poor recall for the minority class.

Answer

Apply PCA and visualize the first two principal components

Answer

Compute pairwise correlation matrix among all features

Answer

Impute missing values using mean imputation

Question 12

After loading a dataset into a pandas DataFrame, a data scientist runs df.info() and sees that a column 'income' has object dtype. What does this indicate, and what EDA step should be taken?

Accepted Answer

The column contains strings; convert to numeric using pd.to_numeric() and investigate non-convertible values.. In pandas, an 'object' dtype column typically means the column contains Python string objects (or mixed types), not numeric values. For an 'income' column, this usually indicates the data was read as strings — possibly due to currency symbols, commas, or non-numeric entries. The correct EDA step is to convert with pd.to_numeric() (using errors='coerce' to identify non-convertible values) and then investigate those values (e.g., 'N/A', '$50,000', blanks) before deciding on imputation or cleaning.

Answer

The column has missing values; impute them.

Answer

Normalize the column to a 0-1 range.

Answer

The column is already numeric; proceed.

Question 13

During EDA, a data scientist notices that a numeric feature 'age' has values ranging from 0 to 150, but expects adult ages between 18-100. Which TWO steps should the scientist take to investigate?

Accepted Answer

Compute summary statistics (min, max, percentiles). Option B is correct because computing summary statistics such as min, max, and percentiles (e.g., with pandas describe() or quantile()) quantifies the distribution of 'age' and reveals how many values fall outside the expected 18-100 range, which is the proper first step in EDA before deciding on any treatment. Option E is correct because a box plot visualizes the median, interquartile range, and points beyond the whiskers (typically 1.5*IQR), directly exposing outliers such as ages near 0 or 150 and showing whether they are isolated or systematic. Option A is not appropriate because deleting rows with age > 100 is a premature data-destruction step; the scientist should first investigate whether those values are data-entry errors, unit issues, or legitimate cases before removing anything. Option C is not appropriate because a log transformation is a modeling/preprocessing technique for skewed distributions and does nothing to diagnose why ages of 0 or 150 exist. Option D is not appropriate because imputing out-of-range ages with the mean would mask the anomaly and distort the distribution before the cause of the invalid values has been investigated.

Answer

Remove all rows with age > 100

Answer

Apply log transformation to normalize the distribution

Answer

Impute age values outside 18-100 with the mean

Question 14

A data engineering team is designing a data pipeline that processes streaming data from Amazon Kinesis Data Streams using AWS Lambda. The team notices that some records are being processed multiple times (duplicates). Which TWO steps should the team take to ensure exactly-once processing?

Accepted Answer

Design the Lambda function to be idempotent.. Option A is correct because Lambda's Kinesis event source mapping is at-least-once by design, so the function must be idempotent to make repeated deliveries of the same record produce the same result without side effects. Option B is correct because tracking a unique record identifier (e.g., Kinesis sequence number or a producer-supplied ID) in an external store like DynamoDB lets the function detect and skip records it has already processed, which is the standard pattern for achieving exactly-once semantics on top of at-least-once delivery. Option C is not correct because increasing batch size only changes invocation frequency and batching, not the at-least-once delivery guarantee, so duplicates can still occur. Option D is not correct because KPL provides reliable, ordered delivery with aggregation and retries, but it does not guarantee exactly-once delivery to consumers. Option E is not correct because disabling retries would cause data loss on transient failures and does not prevent duplicates from Kinesis checkpointing behavior.

Answer

Increase the batch size to reduce the number of invocations.

Answer

Use Kinesis Producer Library (KPL) to guarantee exactly-once delivery.

Answer

Disable retries on the Lambda function.

Question 15

A machine learning team is analyzing a dataset with a target variable that is highly imbalanced (99% negative class, 1% positive class). They want to understand the distribution and relationships before modeling. Which exploratory data analysis technique is most appropriate to visualize the imbalance and guide resampling strategy?

Accepted Answer

Bar chart of class frequencies and a correlation heatmap. A bar chart of class frequencies clearly visualizes the imbalance (99% negative vs 1% positive), and a correlation heatmap helps identify which features are correlated with the target, guiding resampling strategy. Option A is wrong because a confusion matrix is used for evaluating model predictions, not for initial exploratory data analysis of class imbalance. Option B is wrong because a scatterplot matrix is designed to visualize relationships between continuous variables and can be overwhelming with many features; it does not directly highlight the class imbalance. Option C is wrong because box plots grouped by target class show feature distributions across classes but do not explicitly quantify the imbalance ratio itself.

Answer

Confusion matrix on a sample of the data

Answer

Scatterplot matrix of all features colored by class

Answer

Box plots of each feature grouped by the target class

Question 16

A machine learning team is training a deep learning model on Amazon SageMaker and notices that the training loss is decreasing but the validation loss is increasing. What is the most likely cause?

Accepted Answer

Overfitting the training data. When training loss continues to decrease while validation loss increases, the model is memorizing the training data rather than learning generalizable patterns. This is the classic symptom of overfitting, where the model's capacity exceeds what is needed for the underlying data distribution, causing it to fit noise in the training set. In Amazon SageMaker, this can be observed by monitoring the validation loss metric during training jobs.

Answer

Vanishing gradients

Answer

Learning rate is too high

Answer

Underfitting the training data

Question 17

A machine learning engineer needs to prepare a training dataset stored in Amazon S3. The dataset is in CSV format and contains missing values and categorical features. The engineer wants to use Amazon SageMaker Data Wrangler to transform the data and then train a model using SageMaker. The engineer needs to ensure that the transformations are reproducible and can be applied to new data during inference. What should the engineer do?

Accepted Answer

Export the Data Wrangler flow to a SageMaker Pipeline and use the generated preprocessing code in the inference pipeline.. Exporting the Data Wrangler flow to a SageMaker Pipeline generates reusable preprocessing code that can be incorporated into both training and inference pipelines. This ensures that the same transformations are applied consistently, maintaining reproducibility and correctness during inference. The other options either lose the transformation logic or require manual reimplementation, which is not ideal.

Answer

Use Data Wrangler to export the transformed data to S3, then train a model directly on the transformed data without saving the flow.

Answer

Manually recreate the transformations in a Jupyter notebook using pandas and scikit-learn, and use that notebook for both training and inference.

Answer

Use SageMaker Processing jobs to run the Data Wrangler flow on the training data, and then deploy the model without preprocessing for inference.

Question 18

A data science team uses Amazon SageMaker to train models on a large dataset stored in S3. The dataset is 500 GB in CSV format and is updated daily. The team wants to optimize data loading for training jobs to reduce I/O wait time. Which data ingestion strategy is MOST effective?

Accepted Answer

Use SageMaker Pipe input mode to stream data directly from S3.. SageMaker Pipe input mode streams data directly from S3 to the training algorithm without writing to the instance's EBS volume, eliminating disk I/O bottlenecks. This is especially effective for large datasets (500 GB) that are updated daily, as it reduces startup time and avoids the need to download the entire dataset before training begins.

Answer

Use SageMaker File input mode and increase the EBS volume size to 1 TB.

Answer

Convert the CSV files to Parquet format and use File input mode.

Answer

Load the data into an Amazon EFS file system and mount it to the training instance.

Question 19

A data scientist examines a dataset with 100 features and suspects that some features are redundant due to high pairwise correlations. Which EDA technique should the scientist use to systematically identify groups of highly correlated features?

Accepted Answer

Generate a correlation matrix and visualize it as a heatmap.. A correlation matrix heatmap allows systematic identification of groups of highly correlated features by visually highlighting high pairwise correlations. Option B is incorrect because histograms show univariate distributions, not relationships between features. Option C is incorrect because scatter plots for each pair would be time-consuming and not systematic for 100 features. Option D is incorrect because box plots show outliers, not correlations.

Answer

Plot histograms for each feature.

Answer

Create scatter plots for each pair of features.

Answer

Use box plots to identify outliers.

Question 20

A company is using Amazon Kinesis Data Streams to ingest real-time data from millions of IoT devices. The data is consumed by an AWS Lambda function that writes to Amazon S3. The company notices that the Lambda function is throttling and data is being lost. The engineer needs to improve the reliability and scalability of the pipeline. Which solution should the engineer implement?

Accepted Answer

Increase the number of shards in the Kinesis stream and configure the Lambda function to use parallelization factor greater than 1.. To handle high-volume streaming data with Lambda, increasing the number of shards provides more parallelism, and setting the parallelization factor allows multiple Lambda invocations per shard. This scales the processing capacity and reduces throttling. Other options either do not address the scaling issue or replace the architecture in ways that may not meet requirements.

Answer

Increase the Lambda function's memory and timeout settings to allow more time for processing each batch.

Answer

Use Kinesis Data Firehose to deliver data directly to S3, eliminating the need for Lambda processing.

Answer

Replace Kinesis Data Streams with Amazon SQS and have Lambda poll the queue, scaling the number of Lambda functions automatically.