MLS-C01 Practice Test 50 — 15 Questions

Question 1

Refer to the exhibit. A data scientist is trying to run a SageMaker training job using a script that reads data from the S3 bucket 'my-bucket' and writes the model artifact to the same bucket. The training job fails with an access denied error. What is the likely cause?

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

Accepted Answer

The IAM role does not have permission to write to the S3 bucket for the model artifact. The training job fails with an access denied error because the IAM role used by SageMaker lacks the s3:PutObject permission (or equivalent write access) for the S3 bucket 'my-bucket'. While the script reads data from the bucket, writing the model artifact requires explicit write permissions on the same bucket. Without this, SageMaker cannot upload the model artifact, causing the job to fail.

Answer

The IAM role does not have sagemaker:CreateModel permission

Answer

The IAM role does not have s3:ListBucket permission

Answer

The IAM role does not have ec2:DescribeInstances permission

Question 2

A data scientist is training a deep learning model on a GPU instance. The training data is stored in S3 and is 50 GB. To reduce I/O bottlenecks, which storage option should be used to cache the data locally on the instance?

Accepted Answer

Mount an Amazon FSx for Lustre file system linked to the S3 bucket. Amazon FSx for Lustre is a high-performance file system optimized for compute-intensive workloads like deep learning. It can be linked to an S3 bucket, allowing the data to be lazily loaded from S3 and cached on the file system, providing low-latency access to the training data. This reduces I/O bottlenecks by providing high throughput and low latency.

Answer

Attach an Amazon EFS file system to the instance and copy data from S3

Answer

Provision an Amazon EBS io2 volume and copy data from S3 using AWS DataSync

Answer

Use instance store volumes to cache the data from S3

Question 3

A data scientist wants to use Amazon SageMaker to train a deep learning model on a large dataset stored in S3. The training job is expected to take several hours. Which storage option should be used to minimize data loading time and cost?

Accepted Answer

Use Pipe mode to stream data directly from S3 during training. Pipe mode is the correct choice because it streams data directly from S3 into the training algorithm without writing to disk, eliminating the time and cost of copying large datasets to the instance's local storage. This minimizes data loading time (streaming starts immediately) and cost (no EBS volume or additional storage charges), making it ideal for large datasets that take hours to train.

Answer

Attach an Amazon EBS volume with the dataset pre-loaded

Answer

Use File mode to copy data to the training instance's local storage

Answer

Mount an Amazon EFS file system to the training instance

Question 4

A data scientist is using Amazon SageMaker to build a text classification model. The dataset has 100,000 labeled samples and 20 classes. The scientist wants to use a pre-trained BERT model and fine-tune it. Which approach is MOST cost-effective?

Accepted Answer

Fine-tune a pre-trained BERT-base model using a GPU instance.. Fine-tuning a pre-trained BERT-base model on a GPU instance is the most cost-effective approach because transfer learning leverages the pre-trained model's existing language understanding, requiring far less compute and data than training from scratch. BERT-base has 110M parameters and is well-suited for text classification with 100,000 samples across 20 classes. GPU instances accelerate the fine-tuning process significantly compared to CPU, reducing overall cost.

Answer

Train a BERT model from scratch using a larger instance.

Answer

Use a pre-trained BERT-large model with a larger instance.

Answer

Train a CNN model from scratch using CPU instances.

Question 5

A data scientist is using Amazon SageMaker to train a linear learner model for regression. After reviewing the training logs, the data scientist notices that the loss is not decreasing and remains high. The learning rate is set to 0.01. The data is normalized. What should the data scientist do to improve convergence?

Accepted Answer

Try different learning rates, such as 0.001 or 0.1.. When loss is not decreasing and remains high despite normalized data, the most likely culprit is an inappropriate learning rate. A learning rate of 0.01 may be too high (causing divergence/oscillation) or too low (causing slow convergence). Trying values like 0.001 or 0.1 is the standard diagnostic step to find a rate that allows the optimizer to descend the loss surface effectively.

Answer

Normalize the data again.

Answer

Reduce the mini-batch size.

Answer

Increase the number of epochs.

Question 6

A company uses AWS Glue to run ETL jobs that transform data from Amazon RDS for MySQL to Amazon S3. The current job runs daily and takes 3 hours to process 100 GB of data. The company expects data volume to grow 10x in the next year. They need to reduce job runtime and cost. Which approach should they take?

Accepted Answer

Use parallel reads with pushdown predicates in the Glue job's source connection, and write the output in columnar format (Parquet) partitioned by date.. Using parallel reads with pushdown predicates reduces the amount of data transferred from RDS to Glue by filtering at the database level, which lowers extraction time and load on the source. Writing output in columnar format (Parquet) reduces storage size and improves query performance for downstream analytics. Partitioning by date enables efficient pruning. Option A is incorrect because S3 Select is used for server-side filtering of data already in S3, not for tuning extraction from RDS. Option C is incorrect because increasing DPUs alone does not solve the bottleneck from the source database; pushdown predicates are more effective. Option D is incorrect because Redshift Spectrum is used for querying data in S3, not for performing transformations in ETL jobs.

Answer

Use S3 Select with Glue to filter data before transformation.

Answer

Increase the number of Glue DPUs to 100 and enable job bookmarking.

Answer

Use Amazon Redshift Spectrum to perform transformations in place on S3.

Question 7

A company uses Amazon SageMaker to train a model and wants to track metrics like loss and accuracy in real-time. Which SageMaker feature should be used?

Accepted Answer

SageMaker metrics and CloudWatch dashboards. SageMaker's built-in metrics integration with CloudWatch allows real-time tracking of training metrics such as loss and accuracy. Option A (SageMaker Model Monitor) is for monitoring inference quality and data drift after deployment, not real-time training metrics. Option C (SageMaker Experiments) is for organizing, tracking, and comparing different experiment runs, but it does not provide real-time streaming of metrics during training. Option D (SageMaker Debugger) is used for debugging training issues like gradient anomalies, not for real-time metric tracking.

Answer

SageMaker Model Monitor

Answer

SageMaker Experiments

Answer

SageMaker Debugger

Question 8

A team is training a linear regression model to predict house prices. After training, they observe that the model has high bias (underfitting). Which action is most likely to reduce bias?

Accepted Answer

Add more relevant features and increase model complexity.. High bias (underfitting) means the model is too simple to capture the underlying patterns in the data. Adding more relevant features and increasing model complexity (e.g., using polynomial features or more interaction terms) gives the linear regression model greater capacity to fit the training data, directly reducing bias. This aligns with the bias-variance tradeoff, where increasing complexity lowers bias at the cost of potentially increasing variance.

Answer

Increase the regularization strength.

Answer

Reduce the amount of training data.

Answer

Decrease the number of model parameters.

Question 9

Match each AWS security service to its function in ML.

Accepted Answer

AWS KMS: Manages encryption keys for securing ML model data at rest and in transit. In ML, AWS KMS handles encryption keys, IAM manages access, CloudTrail provides audit logs. Common confusions arise because these services often work together, but each has a distinct primary function.

Answer

AWS KMS: Provides identity-based access control for ML resources

Answer

AWS IAM: Encrypts data during ML processing

Answer

AWS CloudTrail: Monitors and enforces compliance rules for ML

Question 10

Match each SageMaker optimization technique to its description.

Accepted Answer

Managed Spot Training: Uses EC2 Spot instances to reduce training cost. Managed Spot Training reduces cost via spot instances; Data Parallelism speeds training by distributing data; Automatic Model Tuning finds optimal hyperparameters; Compilation (Neo) optimizes for inference hardware. Common confusions include swapping definitions between these techniques.

Answer

Managed Spot Training: Optimizes model for inference on edge devices

Answer

SageMaker Data Parallelism: Automatically searches hyperparameter space to optimize model

Question 11

A data scientist is analyzing a dataset with many missing values. The scientist wants to decide on an imputation strategy. Which THREE considerations are important for choosing the imputation method?

Accepted Answer

The mechanism of missingness (MCAR, MAR, MNAR).. Option A is correct because the missingness mechanism (MCAR, MAR, or MNAR) determines whether simple imputation like mean/mode is unbiased or whether model-based/multiple imputation is required; MNAR in particular needs explicit modeling of the missingness. Option C is correct because the proportion of missing values per feature dictates feasibility: low percentages tolerate mean/median/mode imputation, while high percentages (e.g., >40-50%) may warrant dropping the feature or using iterative/multiple imputation. Option D is correct because feature distribution affects the choice of statistic: mean imputation is reasonable for roughly normal data, whereas median or model-based imputation is preferred for skewed distributions to avoid distorting the central tendency. Option B is not a primary driver of imputation method choice; class imbalance concerns modeling/resampling, not how to fill missing cells. Option E is also not a direct criterion, since random-forest feature importance is a downstream modeling output and does not itself determine an appropriate imputation strategy.

Answer

The class imbalance of the target variable.

Answer

The feature importance according to a random forest model.

Question 12

A data scientist is performing exploratory data analysis on a dataset with both numerical and categorical features. The scientist wants to visualize the pairwise relationships between numerical features and also see the distribution of each feature. Which type of plot should the scientist use?

Accepted Answer

Pair plot (scatter matrix) with histograms on the diagonal.. A pair plot (scatter matrix) plots pairwise scatterplots for numerical features and can place histograms (or KDEs) on the diagonal to show each feature's distribution. This single visualization satisfies both requirements: pairwise relationships between numerical features and univariate distributions. Box plots, heatmaps, and correlation matrices each address only part of the requirement.

Answer

Box plot for each feature.

Answer

Heatmap of the correlation matrix.

Answer

Correlation matrix with numbers.

Question 13

A data scientist is training a binary classification model using SageMaker XGBoost and notices that training loss decreases but validation loss increases after a few epochs. Which action should the data scientist take to address this issue?

Accepted Answer

Set early stopping based on validation loss. The increasing validation loss while training loss decreases is a classic sign of overfitting. Setting early stopping based on validation loss halts training when the validation loss stops improving, preventing the model from memorizing noise in the training data. SageMaker XGBoost's `early_stopping_rounds` parameter monitors the evaluation metric on the validation set and stops training if no improvement is seen for a specified number of rounds.

Answer

Increase the number of rounds

Answer

Increase the learning rate

Answer

Increase the maximum tree depth

Question 14

A data scientist is building a binary classification model to predict whether a customer will subscribe to a service. The dataset contains 20 features, including categorical variables with high cardinality (e.g., zip code with 10,000 unique values). The scientist uses a logistic regression model and obtains a training AUC of 0.85 and a test AUC of 0.60. The scientist suspects overfitting due to high cardinality features. Which approach should the scientist use to address this issue?

Accepted Answer

Apply target encoding with smoothing to the zip code feature. Target encoding with smoothing replaces each high-cardinality category with a smoothed statistic (typically the mean of the target) that incorporates a prior, which reduces the overfitting that occurs when raw target means are computed on small groups. This preserves the predictive signal in zip code while regularizing categories with few observations, addressing the train/test AUC gap.

Answer

Apply label encoding to the zip code feature

Answer

Remove the zip code feature entirely

Answer

Apply one-hot encoding to the zip code feature

Question 15

A data scientist is training a deep learning model for image segmentation using a U-Net architecture. The model overfits severely. The scientist tries L2 regularization, dropout, and data augmentation, but validation loss remains high while training loss approaches zero. Which additional strategy is most likely to reduce overfitting?

Accepted Answer

Implement early stopping based on validation loss. Early stopping monitors validation loss and halts training when it stops improving, directly addressing overfitting by preventing the model from memorizing noise after it has learned generalizable features. Since the training loss is near zero but validation loss remains high, the model has already started overfitting, and early stopping can cut training at the point just before overfitting worsens.

Answer

Increase the batch size

Answer

Use a larger learning rate

Answer

Add more convolutional layers to increase model capacity