MLS-C01 Practice Test 37 — 15 Questions

Question 1

A company uses a SageMaker endpoint for real-time inference. They need to ensure high availability during deployment updates. Which THREE steps achieve this? (Choose 3)

Accepted Answer

Use blue/green deployment with a new endpoint configuration. Option B is correct because blue/green deployment creates a new endpoint configuration and fleet, then shifts traffic only after the new instances are healthy, so the old endpoint keeps serving during the update and rollback is possible. Option C is correct because configuring multiple instances behind the endpoint provides redundancy across Availability Zones, so the loss of one instance does not take down real-time inference. Option E is correct because SageMaker's Canary and Linear traffic shifting modes gradually move a percentage of invocations to the new variant while the old variant continues serving, maintaining availability throughout the update. Option A is wrong because a single instance is a single point of failure and cannot provide high availability. Option D is wrong because deleting the old endpoint before creating the new one causes downtime, which is the opposite of high availability.

Answer

Use a single instance to save costs

Answer

Delete the old endpoint before creating the new one

Question 2

A data scientist is analyzing a dataset with a binary target variable. The dataset has 50,000 rows and 200 features. The data scientist wants to identify which features are most predictive. Which TWO methods are appropriate for feature selection during EDA?

Accepted Answer

Mutual information. Mutual information (A) is correct because it measures the dependency between each feature and the binary target, capturing both linear and non-linear relationships, which makes it well-suited for ranking the predictive power of 200 features during EDA. The chi-squared test (B) is also correct because it evaluates the statistical independence between categorical (or discretized) features and a binary target, directly indicating which features are most associated with the target. LSTM neural networks (C) are sequence models for time-series or sequential data and are not a feature-selection method for tabular EDA. K-means clustering (D) is an unsupervised algorithm that groups rows by similarity and does not use the target variable, so it cannot rank feature predictiveness. PCA (E) is a dimensionality-reduction technique that creates new uncorrelated components rather than identifying which original features are most predictive of the target.

Answer

LSTM neural network

Answer

K-means clustering

Answer

Principal Component Analysis (PCA)

Question 3

A machine learning engineer is analyzing a text classification dataset with 50,000 documents. Which EDA step is most important to understand the vocabulary size and frequency distribution?

Accepted Answer

Plot frequency of each word in a bar chart. Plotting the frequency of each word in a bar chart directly visualizes the vocabulary size and the frequency distribution — you can see how many unique words exist, which words dominate, and whether the distribution follows Zipf's law. This is the most direct EDA step for understanding vocabulary characteristics in a text corpus.

Answer

Compute TF-IDF matrix

Answer

Generate bigram collocations

Answer

Plot histogram of document lengths

Question 4

During exploratory data analysis, a data scientist notices that a feature has a highly skewed distribution. Which transformation is most likely to make the distribution approximately normal?

Accepted Answer

Log transformation. A log transformation compresses the right tail of a right-skewed distribution, pulling in extreme high values and making the distribution more symmetric and approximately normal. This is especially effective for positively skewed data that spans several orders of magnitude, such as income or population counts. Min-max scaling and standardization only linearly rescale the data, preserving the original skewness, while one-hot encoding is for categorical variables and does not apply to continuous skewed features.

Answer

Min-max scaling

Answer

One-hot encoding

Answer

Standardization (z-score)

Question 5

A company is deploying a machine learning model using SageMaker. The model is a PyTorch model that requires GPU for inference. The company wants to minimize costs while ensuring low latency. Which instance type should be used for the SageMaker endpoint?

Accepted Answer

ml.g4dn.xlarge. Ml.g4dn.xlarge is a GPU instance with an NVIDIA T4, optimized for inference, and is more cost-effective than ml.p3.2xlarge while still providing low latency for PyTorch models. Option A (ml.m5.large) is wrong as it is a CPU instance without GPU support. Option B (ml.p3.2xlarge) is wrong because although it has a GPU, it is more expensive and not necessary for low-latency inference; ml.g4dn.xlarge offers similar performance at lower cost. Option C (ml.c5.2xlarge) is also a CPU instance and unsuitable.

Answer

ml.m5.large

Answer

ml.p3.2xlarge

Answer

ml.c5.2xlarge

Question 6

A data scientist is working with a dataset that contains geolocation coordinates (latitude and longitude) and timestamps. The scientist wants to visualize the data to check for spatial and temporal patterns. Which TWO AWS services can be used for this visualization?

Accepted Answer

Amazon SageMaker Data Wrangler. Amazon SageMaker Data Wrangler (option B) is correct because it provides built-in data visualization and analysis capabilities, including scatter plots and histograms, that let a data scientist explore spatial (latitude/longitude) and temporal (timestamp) patterns during data preparation. Amazon QuickSight (option D) is correct because it is a BI visualization service that supports geospatial charts (e.g., point maps using latitude/longitude) and time-series visuals with date fields, making it suitable for examining spatial and temporal trends. Amazon Comprehend (option A) is a natural-language-processing service for text, not geospatial or temporal visualization. Amazon Rekognition (option C) performs image and video analysis, which is unrelated to coordinate/timestamp data. AWS Glue (option E) is a serverless ETL and data catalog service; it can prepare data but does not itself provide the requested visualization.

Answer

Amazon Comprehend

Answer

Amazon Rekognition

Answer

AWS Glue

Question 7

A data scientist needs to detect outliers in a dataset with multiple features that follow different distributions. Which method is most robust for multivariate outlier detection?

Accepted Answer

Isolation Forest. Isolation Forest is an ensemble method that isolates anomalies effectively in high-dimensional spaces without assuming any specific distribution. Option A is wrong because Z-score assumes a normal distribution. Option B is wrong because IQR is univariate and does not capture multivariate interactions. Option C is wrong because DBSCAN is primarily a clustering algorithm and is not specifically designed for outlier detection, though it can identify outliers as noise; however, Isolation Forest is more robust for this purpose.

Answer

Z-score threshold

Answer

Interquartile range (IQR)

Answer

DBSCAN clustering

Question 8

A data scientist is using Amazon SageMaker Debugger to monitor training. Which THREE types of issues can Debugger monitor?

Accepted Answer

Poor weight initialization. Amazon SageMaker Debugger monitors training jobs by collecting tensors and metrics and evaluating them against built-in rules, and it can detect poor weight initialization (B) because rules such as dead_relu or saturated_activation flag unhealthy initial activation/weight distributions early in training. It detects overfitting (D) via rules like overfit that compare training loss against validation loss to signal a widening generalization gap. It detects vanishing gradients (E) through rules such as vanishing_gradient that watch gradient magnitudes across layers and alert when they shrink toward zero. The unmarked options do not belong: hardware failures (A) are infrastructure-level events handled by SageMaker job status, CloudWatch, and retry mechanisms rather than Debugger's tensor-analysis rules, and data drift (C) is a model-monitoring concern addressed by SageMaker Model Monitor, not by Debugger during training.

Answer

Hardware failures

Answer

Data drift

Question 9

A data scientist is exploring a dataset with categorical variables. Which TWO EDA techniques are appropriate for understanding the relationship between a categorical feature and a continuous target? (Choose TWO.)

Accepted Answer

Violin plots. Violin plots (B) are correct because they combine a box plot with a kernel density estimate, showing the full distribution of the continuous target within each category, which reveals differences in median, spread, skewness, and multimodality across categories. Side-by-side box plots (E) are also correct because they display the median, quartiles, and outliers of the continuous target for each categorical group, making group-to-group comparisons of central tendency and dispersion straightforward. A correlation matrix (A) is inappropriate because Pearson/Spearman correlation requires numeric variables and cannot directly encode a nominal categorical feature. A scatter plot with the categorical variable on the x-axis (C) is not ideal since the categorical axis has no meaningful numeric spacing, so it degenerates into vertical strips and is better replaced by jittered or violin/box representations. A bar chart of category counts (D) only shows the frequency of each category and says nothing about the continuous target's distribution or relationship with the categorical feature.

Answer

Correlation matrix

Answer

Scatter plot with categorical variable on x-axis

Answer

Bar chart of category counts

Question 10

Refer to the exhibit. An administrator has attached this IAM policy to a user. The user tries to start a SageMaker training job that uses a custom Docker image from Amazon ECR. The training job fails with an access denied error. What is the MOST likely reason?

Accepted Answer

The policy is missing ecr:GetDownloadUrlForLayer and ecr:BatchGetImage.. The IAM policy grants sagemaker:* and s3:* actions but does not include the specific ECR permissions required to pull a custom Docker image. When SageMaker launches a training job with a custom image, it needs ecr:GetDownloadUrlForLayer and ecr:BatchGetImage to authenticate and download the image layers from Amazon ECR. Without these permissions, the training job fails with an access denied error.

Answer

The s3:* action is too permissive and should be scoped.

Answer

The iam:PassRole permission is missing the SageMaker service principal.

Answer

The sagemaker:* action should be restricted to specific resources.

Question 11

A company is building a fraud detection model. The dataset is highly imbalanced (99% legitimate, 1% fraud). The data scientist trains a model using Amazon SageMaker's built-in XGBoost algorithm. The model achieves 99% accuracy but only catches 10% of fraud cases. Which technique should the data scientist apply to improve recall for the minority class?

Accepted Answer

Set the scale_pos_weight hyperparameter in XGBoost.. Setting the scale_pos_weight hyperparameter in XGBoost adjusts the weight of the positive (minority) class during training, effectively penalizing misclassifications of fraud cases more heavily. This directly addresses the class imbalance by forcing the model to focus on the minority class, which improves recall without altering the dataset distribution. The current 99% accuracy with only 10% fraud recall indicates the model is biased toward the majority class, and scale_pos_weight is the most direct and efficient fix within XGBoost.

Answer

Use random under-sampling of the majority class.

Answer

Use mean squared error as the objective function.

Answer

Use SMOTE to oversample the minority class.

Question 12

A data analyst is exploring a dataset and wants to identify outliers in a numerical feature. Which visualization technique is most effective for detecting outliers?

Accepted Answer

Box plot. A box plot is specifically designed to visualize the distribution of a numerical feature and highlight outliers using the interquartile range (IQR) rule — points beyond 1.5×IQR from the quartiles are plotted as individual outliers. It provides an immediate, quantitative view of outliers without needing additional computation. This makes it the most effective single visualization for outlier detection.

Answer

Line chart

Answer

Histogram

Answer

Scatter plot

Question 13

A company is building a binary classification model to predict customer churn. The dataset is highly imbalanced (95% non-churn, 5% churn). The data scientist uses SMOTE to oversample the minority class. After training a logistic regression model, the recall for the churn class is 0.80, but the precision is only 0.10. Which action would MOST likely improve precision without significantly harming recall?

Accepted Answer

Increase the classification threshold for the positive class. Increasing the classification threshold for the positive class (churn) means the model will predict churn only when its estimated probability is higher, which reduces false positives and thus improves precision. Since recall is currently high (0.80), there is room to trade some recall for precision; a higher threshold will filter out weaker positive predictions. This directly addresses the precision-recall tradeoff without altering the underlying model or data.

Answer

Use random oversampling instead of SMOTE

Answer

Reduce the number of features in the model

Answer

Decrease the classification threshold for the positive class

Question 14

A data scientist is using Amazon SageMaker to train a linear regression model. The target variable is right-skewed. Which transformation should the data scientist apply to the target variable to improve model performance?

Accepted Answer

Log transformation. A right-skewed target variable violates the normality assumption of linear regression and can cause heteroscedasticity and unstable coefficient estimates. Applying a log transformation compresses the long right tail, making the distribution more symmetric and bringing the relationship closer to linear, which improves model fit and prediction accuracy. This is a standard preprocessing step for skewed continuous targets.

Answer

Min-max scaling

Answer

One-hot encoding

Answer

Principal Component Analysis (PCA)

Question 15

A data scientist is using SageMaker to train a deep learning model for image classification. The training job is taking too long. Which approach can reduce training time?

Accepted Answer

Use SageMaker's distributed data parallelism. SageMaker's distributed data parallelism splits the training data across multiple GPUs or instances, allowing each worker to process a different subset of the data simultaneously. This reduces the wall-clock time per epoch by parallelizing the computation, which directly addresses the 'taking too long' issue for deep learning image classification models.

Answer

Use SageMaker Neo to compile the model

Answer

Increase the number of epochs

Answer

Use a smaller image size