MLS-C01 Practice Exam 70 — 20 Questions

Question 1

A machine learning engineer is deploying a model to an Amazon SageMaker endpoint for real-time inference. The model requires a preprocessing step that involves tokenizing text and converting it to a numerical format. To minimize latency, where should the preprocessing logic be implemented?

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

Accepted Answer

Inside the SageMaker inference container using the inference.py script. To minimize latency, it's best to include the preprocessing logic inside the inference container that serves the model. This avoids additional network calls to separate preprocessing services.

Answer

Using Amazon SageMaker batch transform

Answer

As a separate AWS Lambda function called before the endpoint

Answer

On the client side before sending the request

Question 2

A data scientist is training a binary classification model on an imbalanced dataset where the positive class represents 5% of the data. Which metric is most appropriate for evaluating model performance?

Accepted Answer

AUC-ROC. AUC-ROC evaluates the model's ability to rank positive instances above negatives across all classification thresholds, making it robust to class imbalance. With only 5% positives, accuracy is misleading because a trivial majority-class predictor achieves 95% accuracy. AUC-ROC provides a threshold-independent measure of discriminative power.

Answer

Accuracy

Answer

Root Mean Squared Error (RMSE)

Answer

R-squared

Question 3

A data scientist is training a binary classifier on an imbalanced dataset (95% negative, 5% positive). The model achieves 99% accuracy but only correctly identifies 2% of the positive samples. Which metric should the data scientist focus on to improve the model's performance?

Accepted Answer

Recall. (Recall). Recall measures the proportion of actual positive samples correctly identified, which is critical for an imbalanced dataset where the model fails to detect positives. Option A (Precision) is not the primary focus because it measures the accuracy of positive predictions, not the ability to find all positives. Option B (RMSE) is a regression metric, not suitable for binary classification. Option D (Accuracy) is misleading because a model can achieve high accuracy by simply predicting the majority class, as seen here.

Answer

Precision

Answer

RMSE

Answer

Accuracy

Question 4

A data scientist is training a deep learning model on Amazon SageMaker and notices that training is taking much longer than expected. The training job uses a single GPU instance. The model is a large transformer with millions of parameters. Which change would most likely reduce training time?

Accepted Answer

Use SageMaker's distributed data parallelism with multiple GPU instances. Distributed data parallelism in SageMaker splits the training dataset across multiple GPU instances, each computing gradients on its own shard and then synchronizing via all-reduce to update the model. This linearly scales throughput for large transformer models whose training time is dominated by forward and backward passes over massive datasets. Since the model already fits on a single GPU, data parallelism is the appropriate scaling strategy rather than model parallelism.

Answer

Reduce the batch size to fit in memory

Answer

Use a smaller instance type

Answer

Switch to a CPU instance

Question 5

A data scientist is using Amazon SageMaker to train a linear learner model for a binary classification problem. The dataset is highly imbalanced, with only 2% positive examples. The model currently predicts the majority class for all instances. Which two actions should the data scientist take to improve the model's ability to detect the positive class? (Choose two.)

Accepted Answer

Use the 'balanced' value for the positive_example_weight_mult parameter.. SageMaker Linear Learner provides the positive_example_weight_mult parameter to handle imbalanced binary classification. Setting it to a value greater than 1 or to 'balanced' increases the weight of positive examples, making the model more sensitive to the minority class. This directly counteracts the bias toward the majority class and improves the model's ability to detect positive instances. Other hyperparameters like learning rate or feature reduction do not address the underlying class imbalance.

Answer

Set the predictor_type to 'regressor' instead of 'binary_classifier'.

Answer

Increase the value of the learning_rate hyperparameter.

Answer

Reduce the number of features using PCA before training.

Question 6

A data analyst is using Amazon SageMaker Data Wrangler to explore a dataset stored in Amazon S3. The analyst wants to quickly identify missing values, outliers, and the distribution of each feature before building a model. Which Data Wrangler feature should the analyst use to obtain this information in a single view?

Accepted Answer

Data quality and insights report. The data quality and insights report in SageMaker Data Wrangler is specifically designed to generate a comprehensive EDA summary, including missing values, outliers, and feature distributions, in one view. Other options are either for bias detection, post-modeling evaluation, or feature importance, none of which provide the requested broad data overview.

Answer

Bias report

Answer

Feature importance plot

Answer

Confusion matrix

Question 7

A machine learning engineer has trained a model using Amazon SageMaker and wants to deploy it to a real-time endpoint. The model artifacts are stored in Amazon S3. The engineer needs to create a SageMaker model, configure an endpoint configuration, and then create the endpoint. Which sequence of API calls should the engineer use to deploy the model?

Accepted Answer

CreateModel, CreateEndpointConfig, CreateEndpoint. Deploying a model to a SageMaker endpoint involves three steps: creating a model that points to the model artifacts and inference code, creating an endpoint configuration that specifies the instance type and count, and creating the endpoint using that configuration. This sequence ensures that the necessary resources are available before the endpoint is launched, allowing for a successful deployment.

Answer

CreateModel, CreateEndpoint, CreateEndpointConfig

Answer

CreateEndpoint, CreateModel, CreateEndpointConfig

Answer

CreateEndpointConfig, CreateModel, CreateEndpoint

Question 8

A machine learning engineer is training a linear regression model on a dataset with 20 features. The training loss decreases but the validation loss increases after a few epochs. The engineer wants to apply a regularization technique that encourages sparsity in the feature coefficients, effectively performing feature selection. Which regularization method should be used?

Accepted Answer

L1 regularization (Lasso). The correct answer is L1 regularization (Lasso) because it induces sparsity by driving some coefficients to zero, effectively selecting a subset of features. This addresses the overfitting issue and meets the goal of feature selection. L2 regularization only shrinks coefficients, dropout is for neural networks, and early stopping does not affect model sparsity.

Answer

L2 regularization (Ridge)

Answer

Early stopping

Answer

Dropout

Question 9

A data scientist is examining a SageMaker training dataset and suspects that a feature's missing values are not random but depend on the target class. They want to determine whether the missingness itself carries predictive signal before deciding how to impute. Which EDA step BEST supports this decision?

Accepted Answer

Segment the data by the missing-value indicator and compare the target distribution and other feature statistics across the two groups.. Comparing target rates and feature statistics between rows where the value is missing and rows where it is present tests whether missingness is informative. If the two groups differ, the missing-value indicator should be retained as a feature and imputation chosen accordingly. Mean imputation, listwise deletion, and KNN imputation all mask or remove the pattern rather than measuring it.

Answer

Apply K-nearest neighbors imputation using only the features with the highest variance and retrain the model.

Answer

Replace all missing values with the column mean and then compute the correlation matrix of the imputed features.

Answer

Drop every row containing a missing value and then profile the remaining complete cases.

Question 10

A data scientist is analyzing a dataset with missing values in several columns. She wants to visualize the pattern of missingness to decide on an imputation strategy. Which visualization should she create using Amazon SageMaker Data Wrangler?

Accepted Answer

Missing value heatmap. SageMaker Data Wrangler includes a 'Missing Value' analysis that produces a heatmap-like visualization, showing which cells are missing. This helps identify patterns such as missingness correlated across columns or rows. Understanding these patterns informs whether to use simple imputation, model-based imputation, or drop missing data. Other visualizations like histograms, scatter plots, and box plots do not depict missingness.

Answer

Scatter plot

Answer

Box plot

Answer

Histogram

Question 11

A machine learning engineer is training a gradient boosting model on Amazon SageMaker for a binary classification task where the positive class is rare (1% of the data). The engineer wants the model to focus more on the minority class during training. Which SageMaker built-in algorithm hyperparameter should the engineer adjust to directly control the weight of the positive class?

Accepted Answer

Set the scale_pos_weight hyperparameter to a high value.. For imbalanced binary classification with SageMaker's built-in XGBoost, the scale_pos_weight hyperparameter is specifically designed to weight the positive class. Setting it to the ratio of negative to positive samples (e.g., 99 for 1% positives) helps the model pay more attention to the minority class. Other hyperparameters like max_depth, eta, and subsample do not directly control class weighting.

Answer

Set the max_depth hyperparameter to a high value.

Answer

Set the subsample hyperparameter to a low value.

Answer

Set the eta hyperparameter to a high value.

Question 12

A machine learning engineer has trained a SageMaker model and wants to deploy it for real-time inference. The model artifacts are stored in Amazon S3, and the engineer needs to ensure that the endpoint can automatically scale based on incoming traffic. The engineer also wants to minimize latency for the inference requests. Which deployment option should the engineer use?

Accepted Answer

SageMaker real-time endpoint with Application Auto Scaling. A real-time endpoint keeps the model loaded and ready, delivering low-latency predictions. Application Auto Scaling dynamically adjusts instance count based on CloudWatch metrics, providing automatic scaling. Together, they address both the latency and scaling needs. Other options either lack real-time capabilities or introduce delays unsuitable for low-latency requirements.

Answer

SageMaker asynchronous inference

Answer

SageMaker serverless inference

Answer

SageMaker batch transform

Question 13

A machine learning engineer is deploying a model to an Amazon SageMaker endpoint for real-time inference. The model requires a preprocessing step that involves tokenizing text and converting it to a numerical format. To minimize latency, where should the preprocessing logic be implemented?

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

Accepted Answer

Deploy the model to a real-time endpoint with a minimum instance count of 2 and auto scaling enabled.. Using a real-time endpoint with a minimum instance count of 2 and auto scaling ensures that at least two instances are always running, eliminating cold starts for those instances. Auto scaling quickly adds more instances when traffic increases, maintaining low latency. This strategy balances cost and performance for variable workloads.

Answer

Deploy the model using SageMaker Serverless Inference.

Answer

Deploy the model using SageMaker Asynchronous Inference.

Answer

Deploy the model to a real-time endpoint with a single instance and no auto scaling.

Question 14

A data scientist is training a linear regression model using Amazon SageMaker's Linear Learner algorithm on a dataset with 50 features. After training, the model's predictions on the validation set are consistently off by a large margin. The scientist suspects that the features have vastly different scales. Which preprocessing step should be applied to the training data to improve the model's performance?

Accepted Answer

Normalize the features to have zero mean and unit variance.. Linear models like Linear Learner rely on gradient descent, which converges faster and more reliably when features are on similar scales. Normalizing features to zero mean and unit variance ensures that the optimization landscape is well-conditioned, preventing features with larger scales from dominating the gradient updates. This leads to better model performance and more accurate predictions.

Answer

Apply one-hot encoding to all numerical features.

Answer

Remove features with low variance.

Answer

Apply principal component analysis (PCA) to reduce dimensionality.

Question 15

A data scientist is using Amazon SageMaker to train a deep learning model for image classification with the built-in Image Classification algorithm. The model's training accuracy is high, but validation accuracy is significantly lower, indicating overfitting. The scientist wants to reduce overfitting without collecting more data. Which two actions should the scientist take? (Choose two.)

Accepted Answer

Increase the dropout_ratio hyperparameter.. Increasing dropout_ratio and weight_decay are both regularization techniques that directly reduce overfitting. Dropout prevents complex co-adaptations, while weight decay penalizes large weights. Together, they improve generalization without requiring additional data. Other options do not specifically target overfitting.

Answer

Increase the number of epochs.

Answer

Increase the batch_size hyperparameter.

Answer

Decrease the learning_rate hyperparameter.

Question 16

A data engineer must catalog a growing collection of JSON files in Amazon S3 so that analysts can query them with Amazon Athena. The files arrive continuously, and new partitions appear daily. The engineer wants the table metadata to stay current automatically with minimal operational effort. Which solution meets these requirements?

Accepted Answer

Run an AWS Glue crawler on a schedule against the S3 prefix and let it update the AWS Glue Data Catalog table.. A scheduled AWS Glue crawler is the standard low-code way to keep S3-backed Athena tables current: it samples the data to infer schema, discovers partition prefixes, and updates the Data Catalog that Athena consults. Manual DDL, custom DynamoDB metadata, or relocating data to Redshift all fail to deliver automatic, low-effort catalog maintenance.

Answer

Load the JSON files into Amazon Redshift using COPY and query Redshift instead of Athena.

Answer

Create an AWS Lambda function that scans the prefix and writes partition metadata to a DynamoDB table for Athena to read.

Answer

Define the table manually in Athena with CREATE TABLE and rely on Athena's schema-on-read to detect new partitions.

Question 17

A machine learning engineer is using Amazon SageMaker to train a model. The training job is configured to use a Spot instance to reduce costs. The engineer wants to ensure that the training job can resume from the last saved checkpoint if the Spot instance is interrupted. Which SageMaker feature should the engineer use?

Accepted Answer

Enable managed spot training and specify an S3 bucket for checkpoints.. Managed spot training is the SageMaker feature that enables training on Spot instances with automatic checkpointing and resumption. By specifying an S3 bucket for checkpoints, the training job can save progress and resume from the last checkpoint after an interruption. Other options do not provide this capability.

Answer

Use SageMaker Debugger to save model checkpoints during training.

Answer

Configure the training job to use a persistent volume for checkpoint storage.

Answer

Set the training job's StoppingCondition to a longer duration to prevent interruption.

Question 18

A data scientist is performing exploratory data analysis on a dataset with a highly imbalanced binary target variable (1% positive). The scientist wants to understand the distribution of a numerical feature 'transaction_amount' across the two target classes. Which two visualizations are most appropriate for this analysis? (Choose two.)

Accepted Answer

Violin plot of 'transaction_amount' grouped by the target class.. To compare the distribution of a numerical feature across two target classes, grouped box plots and violin plots are ideal. They show summary statistics and density, respectively, allowing the scientist to see differences in median, spread, and shape between the positive and negative classes. Scatter plots, pie charts, and ungrouped histograms do not effectively convey this comparison, especially with imbalanced data.

Answer

Scatter plot of 'transaction_amount' versus the target class.

Answer

Pie chart of the target class distribution.

Answer

Histogram of 'transaction_amount' for the entire dataset without grouping.

Question 19

A machine learning engineer is deploying a model to an Amazon SageMaker endpoint for real-time inference. The model requires a preprocessing step that involves tokenizing text and converting it to a numerical format. To minimize latency, where should the preprocessing logic be implemented?

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

Accepted Answer

Create a CloudWatch alarm based on the Model Monitor metrics. SageMaker Model Monitor captures endpoint data and compares it to a baseline to detect drift and quality issues. It publishes metrics to CloudWatch, where alarms can be set to trigger notifications. Together, these provide automated monitoring and alerting. Other options do not directly monitor model performance or provide the required alerting.

Answer

Configure SageMaker Debugger to analyze training job tensors

Answer

Use AWS CloudTrail to log all API calls to the endpoint

Answer

Set up an AWS Lambda function to periodically invoke the endpoint and check responses

Question 20

A data scientist is using Amazon SageMaker to train a custom TensorFlow model with distributed training across multiple GPU instances. The training job uses the SageMaker distributed data parallel library. They observe that the training throughput does not scale linearly with the number of instances; adding more instances yields diminishing returns. Which factor is most likely causing this scalability issue?

Accepted Answer

The model has a large number of parameters, causing high communication overhead during gradient synchronization.. In distributed data parallel training, gradient synchronization across workers introduces communication overhead. For models with many parameters, the all-reduce operation can dominate training time, leading to sublinear scaling. The SageMaker distributed data parallel library mitigates this with communication overlap and optimized algorithms, but large models still face limits. Other factors like data loading are less likely to cause diminishing returns.

Answer

The training dataset is stored in Amazon S3 and not in Amazon FSx for Lustre.

Answer

The learning rate is too high, causing the model to diverge.

Answer

The instances are of different types, causing load imbalance.