MLS-C01 Practice Exam 62 — 20 Questions

Question 1

A machine learning team is using Amazon SageMaker to train a model using a custom Docker container. The training job fails with an error: 'Unable to write to /opt/ml/model'. The container does not have root access. What is the most likely cause?

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

Accepted Answer

The container process does not have write permission to /opt/ml/model. Amazon SageMaker mounts a writable volume at /opt/ml/model inside the training container, but the directory is owned by root with permissions that may not allow non-root users to write. Since the container runs as a non-root user, it lacks the necessary write permission, causing the 'Unable to write to /opt/ml/model' error. The fix is to ensure the Dockerfile creates the directory with appropriate ownership or the training script writes to a different location that is writable, such as /opt/ml/output or /tmp, but the model artifacts must ultimately be saved to /opt/ml/model for SageMaker to capture them.

Answer

The /opt/ml/model directory does not exist in the container

Answer

The container is using an unsupported operating system

Answer

The container does not have internet access

Question 2

A data scientist is using Amazon SageMaker to train a deep learning model for image classification with a large dataset. The training job is taking an excessively long time, and the data scientist wants to optimize the training process without compromising model accuracy. The dataset is stored in Amazon S3, and the training instances are ml.p3.8xlarge. Which two actions should the data scientist take to reduce training time? (Choose two.)

Accepted Answer

Use SageMaker Distributed Training with the Horovod framework to parallelize training across multiple instances.. To reduce training time without sacrificing accuracy, the data scientist should improve data ingestion and parallelize computation. Pipe mode streams data efficiently, avoiding the overhead of downloading large datasets. Distributed training with Horovod splits the workload across multiple instances, leveraging parallel compute to shorten training duration. Both actions directly target the bottlenecks in large-scale deep learning training on SageMaker.

Answer

Reduce the size of the training dataset by randomly sampling 50% of the images.

Answer

Decrease the learning rate to allow the model to converge more smoothly.

Answer

Switch to a smaller instance type, such as ml.m5.large, to reduce cost and increase availability.

Question 3

A financial institution is building a fraud detection model using Amazon SageMaker. The dataset is highly imbalanced, with only 0.5% of transactions labeled as fraudulent. The data scientist trains a binary classification model using the XGBoost algorithm and notices that the model predicts almost all transactions as non-fraudulent, achieving 99.5% accuracy but failing to detect any fraud. Which evaluation metric should the data scientist use to better assess the model's performance in this scenario?

Accepted Answer

Precision-Recall Area Under the Curve (AUC-PR). For highly imbalanced classification problems, precision-recall AUC is the preferred metric because it evaluates the model's performance on the positive (minority) class. It combines precision and recall, showing how well the model identifies fraud without being overwhelmed by the large number of true negatives. This makes it far more informative than accuracy or ROC AUC in this context.

Answer

Accuracy

Answer

Mean Squared Error (MSE)

Answer

Area Under the ROC Curve (AUC-ROC)

Question 4

A financial institution is training a binary fraud detection model on a dataset with 0.2% positive cases. After training an XGBoost model in Amazon SageMaker, the data scientist observes that the model predicts 'not fraud' for every validation sample, achieving 99.8% accuracy. The team wants the model to identify fraud cases while keeping false positives manageable. Which approach should the data scientist take?

Accepted Answer

Use the SageMaker XGBoost built-in algorithm with scale_pos_weight set to the ratio of negative to positive examples.. The core issue is severe class imbalance: with 0.2% positives, a model can achieve high accuracy by always predicting the majority class. The SageMaker XGBoost algorithm supports the scale_pos_weight hyperparameter, which multiplies the gradient of positive examples, effectively rebalancing the loss. This makes the model pay attention to fraud cases, improving recall, and the team can then adjust the classification threshold to control false positives.

Answer

Change the objective to reg:squarederror so the model can output continuous fraud probability scores.

Answer

Increase the number of boosting rounds and lower the learning rate to allow the model to learn minority patterns.

Answer

Apply L2 regularization with a high lambda value to prevent overfitting to the majority class.

Question 5

Which THREE techniques are commonly used to detect outliers in a dataset? (Select THREE.)

Accepted Answer

Interquartile range (IQR). The Interquartile Range (IQR) method (A) is a standard statistical technique that flags outliers as points falling below Q1 − 1.5×IQR or above Q3 + 1.5×IQR, making it a classic univariate outlier detector. The Z-score (D) measures how many standard deviations a value lies from the mean, and points with |Z| above a threshold such as 3 are commonly treated as outliers. Isolation Forest (E) is a dedicated ensemble anomaly-detection algorithm that isolates observations via random partitioning, so outliers require fewer splits and receive shorter path lengths. k-means clustering (B) is a clustering algorithm, not an outlier-detection method, though it can be repurposed indirectly. Principal component analysis (C) is a dimensionality-reduction technique and does not by itself identify outliers.

Answer

k-means clustering

Answer

Principal component analysis (PCA)

Question 6

A machine learning engineer is using Amazon SageMaker to train a model with the built-in Linear Learner algorithm. The engineer wants to perform hyperparameter tuning to improve model performance. Which two hyperparameters are most appropriate to tune for the Linear Learner algorithm? (Choose two.)

Accepted Answer

`wd` (weight decay). The correct answers are `learning_rate` and `wd` (weight decay). For Linear Learner, learning_rate controls the optimization step size, which is fundamental to how well the model converges. wd is the L2 regularization parameter that controls overfitting by penalizing large weights. Both directly influence the model's learned parameters and generalization performance. Other hyperparameters like num_epochs, mini_batch_size, and target_recall are either less critical for performance tuning or serve post-training adjustments.

Answer

`mini_batch_size`

Answer

`num_epochs`

Answer

`target_recall`

Question 7

A data scientist is using Amazon SageMaker to train a model using the built-in XGBoost algorithm. The training job is taking a long time. The data scientist notices that the input data is in CSV format and the training job is using File mode. The data size is 50 GB. What is the BEST way to reduce training time?

Accepted Answer

Switch the input mode to Pipe.. Switching from File mode to Pipe mode streams training data directly from S3 into the algorithm without downloading the full dataset to the training instance's EBS volume first. For a 50 GB CSV dataset, File mode incurs significant download time before training begins, whereas Pipe mode starts training almost immediately and reduces both startup latency and disk space requirements. This is the standard SageMaker optimization for large datasets.

Answer

Use a larger instance type with more vCPUs.

Answer

Convert the data to Parquet format.

Answer

Reduce the number of features in the dataset.

Question 8

A data engineering team is designing a data lake on Amazon S3. The data is ingested from multiple sources in JSON, CSV, and Parquet formats. The team needs to make the data available for analysis using Amazon Athena and Amazon Redshift Spectrum. The team wants to minimize data transformation costs and storage overhead. Which data storage approach should the team use?

Accepted Answer

Convert all data to Apache Parquet before storing in S3. Converting all data to Apache Parquet before storing in S3 minimizes storage overhead and improves query performance. Parquet is a columnar format that provides efficient compression and encoding schemes, reducing storage costs. It is natively supported by Amazon Athena and Redshift Spectrum, enabling fast analytics without on-the-fly conversion. Option B (storing in original format) increases storage costs and can degrade query performance, especially with JSON or CSV. Option C incurs transformation costs each time data is queried, negating any storage benefit. Option A adds unnecessary transformation steps and cluster costs. Therefore, upfront conversion to Parquet is the most cost-effective strategy for this use case.

Answer

Load the data into Amazon Redshift cluster and then unload to S3 in Parquet

Answer

Store the data in its original format in S3 and use Athena to query directly

Answer

Store the data in its original format and use AWS Glue to convert to Parquet when queried

Question 9

A data scientist is training a model on Amazon SageMaker using the built-in Object2Vec algorithm to learn embeddings for a recommendation task. The training dataset consists of user-item interaction pairs. After training, the data scientist wants to evaluate the quality of the learned embeddings. Which evaluation method is most appropriate for assessing the embeddings?

Accepted Answer

Use the embeddings as features to train a downstream classifier and evaluate its accuracy on a held-out set.. Extrinsic evaluation by training a downstream classifier on the embeddings and measuring its accuracy on held-out data is a robust way to assess embedding quality. It tests whether the embeddings capture information that generalizes to new data, which is the ultimate goal for recommendation tasks. Intrinsic metrics like loss or visualization are less directly tied to practical utility.

Answer

Compute the average cosine similarity between random user-item pairs.

Answer

Perform a t-SNE visualization of the embeddings and visually inspect for clusters.

Answer

Measure the training loss of the Object2Vec model on the test dataset.

Question 10

A data scientist is training a deep learning model using Amazon SageMaker and wants to use distributed training to speed up the process. The model is a large transformer with billions of parameters, and the training data is stored in Amazon S3. The data scientist decides to use SageMaker's distributed data parallel library. Which of the following is a key consideration when using this library?

Accepted Answer

The library can be enabled by specifying the distribution parameter with 'smdistributed' and 'dataparallel' in the estimator.. SageMaker's distributed data parallel library is enabled through the estimator's distribution parameter, specifying 'smdistributed' and 'dataparallel'. It does not require Horovod, does not auto-shard data, and while EFA is beneficial, it is not strictly required for all instances.

Answer

The library requires the training script to use the Horovod API for communication.

Answer

The library requires the use of Elastic Fabric Adapter (EFA) for all instance types.

Answer

The library automatically shards the training data across instances, so no additional data partitioning is needed.

Question 11

A data engineer is building a data pipeline for a machine learning project using Amazon SageMaker. The raw data is stored in Amazon S3. Which TWO steps are essential to ensure data privacy and security before training? (Choose TWO.)

Accepted Answer

Encrypt the data at rest using S3 server-side encryption. Option C is correct because encrypting the raw data at rest with S3 server-side encryption (SSE-S3, SSE-KMS, or SSE-C) protects the training data from unauthorized access if the underlying storage is compromised, which is a fundamental data-privacy requirement before feeding data into SageMaker. Option E is correct because running SageMaker jobs inside a VPC and using VPC endpoints (AWS PrivateLink) for S3 keeps traffic between the training containers and S3 on the AWS private network, preventing exposure over the public internet and enforcing network-level isolation. Option A is not essential because restricting access to a single IAM role is only one aspect of least-privilege access and does not by itself guarantee privacy/security of the data at rest or in transit; broader IAM and bucket policies are already implied. Option B is not essential because S3 versioning aids durability and recovery from accidental deletion or overwrites, not data privacy or security. Option D is not essential because S3 Transfer Acceleration only improves upload performance via edge locations and has no bearing on privacy or security.

Answer

Create a bucket policy that restricts access to the data scientist's IAM role only

Answer

Enable versioning on the S3 bucket

Answer

Use S3 Transfer Acceleration for faster uploads

Question 12

A data scientist is building a recommender system using collaborative filtering. The dataset is sparse (99% missing values). Which algorithm is best suited?

Accepted Answer

Matrix Factorization (e.g., SVD). Matrix factorization (e.g., SVD) is best suited for sparse collaborative filtering because it learns latent factors that capture underlying user-item interactions, effectively handling the 99% missing values by generalizing patterns rather than relying on explicit pairwise similarities. Unlike memory-based methods, it decomposes the sparse user-item matrix into lower-dimensional representations, enabling accurate predictions even when most entries are unobserved.

Answer

Random Forest

Answer

K-Nearest Neighbors

Answer

Hidden Markov Model

Question 13

A machine learning engineer is using Amazon SageMaker to train a deep learning model for medical image segmentation. The dataset contains 10,000 images with pixel-level masks. The engineer wants to use the built-in Semantic Segmentation algorithm but finds that the algorithm expects a specific input format. The images and masks are stored in Amazon S3 as PNG files. Which preprocessing step is required to prepare the data for the SageMaker Semantic Segmentation algorithm?

Accepted Answer

Create a JSON label map file that maps class IDs to class names, and organize images and masks in separate folders.. The SageMaker Semantic Segmentation algorithm expects a label map JSON file that defines the mapping between class IDs and class names, including a background class. Images and masks must be in PNG format and organized in separate folders. The training job then references the S3 paths for images and masks. Creating this label map and structure is the essential preprocessing step before launching training.

Answer

Flatten each image and mask into a CSV file with one row per pixel, including x, y coordinates and class labels.

Answer

Convert the images and masks to RecordIO protobuf format using the im2rec tool.

Answer

Convert all images to grayscale and resize them to 256x256 pixels, then store as TFRecord files.

Question 14

A machine learning engineer is training a SageMaker built-in k-means model to segment customers based on behavioral features. The engineer wants to choose the number of clusters and evaluate cluster quality. The training data is unlabeled. Which approach is MOST appropriate for selecting the number of clusters in this scenario?

Accepted Answer

Train models with different k values and compare the average distance of each point to its assigned cluster centroid.. Because clustering is unsupervised, cluster-count selection must rely on intrinsic quality measures. The within-cluster sum of squared distances to centroids is the objective k-means minimizes, so comparing it across several k values, typically with an elbow plot, shows the point of diminishing returns. This approach requires no labels and is directly supported by the SageMaker built-in k-means training output.

Answer

Use the F1 score computed on a held-out validation set to pick the best k.

Answer

Use the model's training accuracy metric reported in the SageMaker training logs.

Answer

Increase k until the number of clusters equals the number of training examples, ensuring every point is its own cluster.

Question 15

Refer to the exhibit. An IAM policy is attached to a SageMaker notebook instance. The data scientist runs a training job that reads from s3://my-bucket/training-data/ and writes to s3://my-bucket/output/. The training job fails with an access denied error. What is the most likely cause?

Accepted Answer

The policy does not allow s3:PutObject on the output location. The training job reads from s3://my-bucket/training-data/ and writes to s3://my-bucket/output/. If the IAM policy grants s3:GetObject on the training prefix but omits s3:PutObject on the output prefix, the job will start successfully (read access works) but fail when it attempts to write model artifacts or output data. That matches the 'access denied' error occurring mid-job rather than at submission time.

Answer

The policy does not allow sagemaker:CreateTrainingJob

Answer

The policy is missing the sagemaker:InvokeEndpoint action

Answer

The policy does not allow s3:GetObject on the training data

Question 16

A data scientist is using Amazon SageMaker to train a model on a dataset stored in Amazon S3. The dataset is in CSV format and is 500 GB in size. The training job is taking a long time to start because SageMaker is downloading the entire dataset to the training instance's storage volume before training begins. The data scientist wants to reduce the startup time and overall training time. Which solution should the data scientist implement?

Accepted Answer

Use SageMaker's Pipe mode with the dataset stored in Amazon S3 and set the input mode to 'Pipe'.. Pipe mode streams data directly from S3 to the training algorithm, avoiding the need to download the entire dataset to local storage before training. This reduces startup time and allows training to begin almost immediately. It also overlaps data loading with computation, potentially reducing overall training time. For a 500 GB dataset, this is the most efficient approach. The data scientist should verify that the chosen algorithm supports Pipe mode.

Answer

Use SageMaker's File mode with the dataset stored in Amazon EFS and mount it to the training instance.

Answer

Use SageMaker's FastFile mode with the dataset stored in Amazon S3 and set the input mode to 'FastFile'.

Answer

Increase the size of the training instance's storage volume to accommodate the entire dataset faster.

Question 17

A data scientist is analyzing a dataset with a time series component. They suspect there is a weekly seasonality. Which technique should they use to confirm this?

Accepted Answer

Compute autocorrelation function (ACF). The autocorrelation function (ACF) measures the correlation between a time series and its lagged values. A significant spike at lag 7 confirms weekly seasonality. Option A (line chart) is subjective and not a definitive test. Option C (Fourier transform) identifies frequency components but is more complex and less direct for confirming seasonality. Option D (moving average) smooths the series and may obscure seasonality.

Answer

Plot the time series line chart

Answer

Perform Fourier transform

Answer

Compute a 7-day moving average

Question 18

A financial institution is building a fraud detection model using Amazon SageMaker. The dataset is highly imbalanced: only 0.5% of transactions are fraudulent. The team wants to evaluate model performance in a way that reflects the business cost of missing a fraudulent transaction, which is 100 times higher than the cost of a false alarm. Which evaluation strategy is MOST appropriate?

Accepted Answer

Use a cost-sensitive metric such as weighted false positive and false negative rates, or optimize expected monetary cost.. The key requirement is to reflect the 100:1 cost asymmetry between false negatives and false positives. A cost-sensitive evaluation metric, such as expected monetary cost or a weighted error rate, directly incorporates that asymmetry and supports threshold selection that minimizes business loss. Accuracy, ROC-AUC, and PR-AUC do not encode the specific cost ratio, so they cannot guarantee alignment with the stated business objective.

Answer

Use area under the ROC curve (AUC-ROC) and select the model with the highest AUC.

Answer

Use accuracy as the primary metric and optimize for it.

Answer

Use precision-recall AUC (PR-AUC) and select the model with the highest PR-AUC.

Question 19

A data scientist is building a binary classifier for loan default prediction. The cost of a false negative (missing a default) is 10 times higher than the cost of a false positive. Which evaluation metric is MOST appropriate?

Accepted Answer

F-beta score with beta=2. The F-beta score with beta=2 is the most appropriate metric because it weights recall (sensitivity) higher than precision, which is critical when false negatives are 10 times more costly than false positives. Beta=2 means recall is considered 2^2 = 4 times more important than precision, directly aligning with the asymmetric cost structure. This allows the model to be tuned to minimize missed defaults, even at the expense of more false alarms.

Answer

Precision

Answer

Accuracy

Answer

Area under the ROC curve

Question 20

An ML engineer is exploring a training dataset for a fraud-detection model. The 'transaction_amount' feature has a heavy right skew and several extreme values. She wants a quick visual comparison of the distribution before and after applying a logarithmic transformation, all inside her Amazon SageMaker Studio notebook. Which approach is MOST appropriate?

Accepted Answer

Use pandas and matplotlib in the notebook to plot overlaid histograms or KDE curves of the raw and log-transformed values.. Overlaid histograms or KDE curves of the raw and log-transformed transaction_amount values directly visualize the effect of the transformation on skew, and pandas with matplotlib is immediately available in a SageMaker Studio notebook. Correlation analysis, Feature Store metrics, and SageMaker Debugger all address different concerns and cannot produce the before/after distribution comparison.

Answer

Compute the Pearson correlation between transaction_amount and the binary fraud label and plot a scatter matrix.

Answer

Enable SageMaker Debugger on a training job and review the tensor distribution summaries it emits.

Answer

Register the feature in Amazon SageMaker Feature Store and inspect the feature's online store metrics.