MLS-C01 Practice Exam 68 — 20 Questions

Question 1

A machine learning engineer is monitoring a deployed model on SageMaker and notices that the prediction latency is increasing over time. The model is a linear regression with a small number of features. Which is the MOST likely cause?

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

Accepted Answer

The inference code has a memory leak. A linear regression model with a small number of features has a tiny memory and compute footprint, so the model itself is not the bottleneck. Increasing prediction latency over time on a long-running SageMaker endpoint is a classic symptom of a memory leak in the inference code — for example, accumulating request data in a global list, not releasing tensors, or leaking file handles. As memory fills, garbage collection and swapping increase, degrading latency progressively.

Answer

The number of features is too large

Answer

The CPU utilization is too low

Answer

The model is overfitting to recent data

Question 2

Which TWO options are valid ways to reduce the amount of data scanned by Amazon Athena queries, thereby reducing cost?

Accepted Answer

Use columnar storage formats like Parquet or ORC. Option A is correct because Athena charges based on the amount of data scanned, and columnar formats such as Parquet or ORC store data by column so that queries reading only a subset of columns scan far less data than row-based formats like CSV or JSON. Option E is correct because partitioning the data by a frequently filtered column (for example, partitioning by date or region) lets Athena use partition pruning to skip entire partitions that do not match the query's WHERE predicate, drastically reducing bytes scanned. Option B is not valid because a LIMIT clause only restricts the number of rows returned after scanning has already occurred; Athena still scans the full underlying data unless the query itself is constrained by partitions or column projection. Option C is not valid because converting data to CSV is a row-based, non-columnar format that typically increases the data scanned compared to Parquet or ORC. Option D is not valid because Athena does not natively support materialized views as a cost-reduction mechanism; while you can create views, they are logical and do not precompute or store results to reduce scanned data.

Answer

Use LIMIT clause in SQL queries

Answer

Convert data to CSV format

Answer

Create materialized views in Athena

Question 3

A data scientist needs to query a 2 TB dataset stored in Amazon S3 using Amazon Athena. The data is in CSV format and is used for exploratory analysis. Queries are currently slow and expensive. Which action will improve query performance and reduce cost?

Accepted Answer

Convert the data to Parquet format and partition by a commonly filtered column.. Converting CSV data to Parquet (a columnar storage format) significantly reduces the amount of data scanned by Athena, as only the columns needed for the query are read. Partitioning by a commonly filtered column (e.g., date or region) further limits the data scanned to only relevant partitions, directly reducing both query cost (Athena charges per TB scanned) and query execution time.

Answer

Convert the data to JSON format to improve compression.

Answer

Increase the number of workers in the Athena query engine.

Answer

Create a composite index on the data using Athena's index feature.

Question 4

A data scientist is deploying a PyTorch model to Amazon SageMaker for real-time inference. The model runs on a large instance but inference latency is too high. Which action is MOST likely to reduce latency without sacrificing accuracy?

Accepted Answer

Compile the model using SageMaker Neo. SageMaker Neo compiles the trained model into an optimized runtime using Apache TVM, applying graph-level optimizations, operator fusion, and memory layout transformations specifically tuned for the target hardware. This reduces inference latency by improving computational efficiency without altering the model's weights or architecture, thus preserving accuracy.

Answer

Switch from a GPU instance to a CPU instance

Answer

Quantize the model weights from FP32 to INT8

Answer

Deploy the model to a multi-model endpoint

Question 5

A company uses Amazon EMR to run Spark jobs on a cluster with 10 core nodes of type r5.xlarge. The jobs are I/O intensive and read large amounts of data from S3. The team notices high network throughput but low CPU utilization. Which configuration change would improve job performance at the same cost?

Accepted Answer

Replace the core nodes with r5d.xlarge instances that have local SSDs.. R5d instances include local NVMe SSDs. These SSDs can be used for caching intermediate data during Spark jobs, reducing the need to read from and write to S3 over the network. This directly addresses the I/O bottleneck and high network throughput observed, improving job performance. Option A is incorrect because moving to general-purpose m5 instances does not provide local SSDs and may not improve I/O. Option B is incorrect because doubling the number of core nodes would increase cost significantly without necessarily solving the I/O issue. Option D is incorrect because spot instances reduce cost but do not inherently improve I/O performance; they may even add instability.

Answer

Change the instance type to m5.xlarge (general purpose) to balance resources.

Answer

Increase the number of core nodes to 20.

Answer

Use spot instances for the core nodes to save cost and reinvest in more nodes.

Question 6

A startup is deploying a machine learning model for real-time recommendation on Amazon SageMaker. The model is a TensorFlow model (1 GB) and the endpoint uses a single ml.c5.2xlarge instance. The inference latency is currently 500 ms per request. The startup expects traffic to increase 10x in the next month. They want to maintain latency under 500 ms. What is the most cost-effective solution?

Accepted Answer

Set up auto-scaling for the endpoint based on average latency or request count. Setting up auto-scaling for the SageMaker endpoint based on average latency or request count allows the endpoint to automatically add instances as traffic increases, maintaining latency under 500 ms. This is cost-effective because it scales out only when needed and scales in during low traffic, avoiding over-provisioning.

Answer

Use SageMaker Batch Transform to process requests in batches

Answer

Switch to a GPU instance type for faster inference

Answer

Upgrade to a larger CPU instance type, such as ml.c5.4xlarge

Question 7

A data scientist is building a regression model to predict house prices. The dataset has 10 features, and the model shows high variance with a low bias. Which technique should the data scientist use to reduce variance?

Accepted Answer

Apply L2 regularization to the model.. L2 regularization (Ridge regression) penalizes large coefficients by adding a squared magnitude term to the loss function, which shrinks the model's weights and reduces variance without substantially increasing bias. This directly addresses the high-variance, low-bias symptom, making the model less sensitive to fluctuations in the training data.

Answer

Increase the depth of decision trees in the ensemble.

Answer

Add more features to the model.

Answer

Reduce the amount of training data.

Question 8

A data scientist is evaluating a regression model. The RMSE on the training set is 2.5, and on the test set is 2.7. The R² on the test set is 0.98. What does this indicate?

Accepted Answer

The model generalizes well with no severe overfitting. The training RMSE (2.5) and test RMSE (2.7) are very close, and the test R² of 0.98 is high, indicating the model explains 98% of the variance on unseen data. A small gap between train and test error is the hallmark of good generalization. There is no evidence of severe overfitting or high bias.

Answer

The model has high bias

Answer

The model is underfitting because R² is too high

Answer

The model is overfitting because RMSE is lower on training data

Question 9

A media company uses SageMaker to deploy a real-time inference endpoint for content recommendation. The model is a PyTorch model that uses GPU. The endpoint is deployed with an ml.p3.2xlarge instance. Over time, the endpoint's latency increases significantly during peak hours. The company has enabled auto scaling based on CPU utilization. However, the latency spikes occur even when CPU utilization is low. The model is stateless and the inference code is efficient. What is the MOST likely cause of the latency spikes?

Accepted Answer

Auto scaling is configured based on CPU utilization, but the bottleneck is GPU utilization. The most likely cause is that auto scaling is based on CPU utilization, but the model uses GPU, and the bottleneck is GPU utilization. During peak hours, GPU utilization may max out while CPU remains low, so auto scaling does not trigger, leading to latency spikes. This is a classic mismatch between the scaling metric and the actual resource bottleneck.

Answer

The model uses stateful processing that accumulates requests

Answer

The inference container has a memory leak that causes gradual slowdown

Answer

The instance type is too small for the model

Question 10

A company is using Amazon SageMaker Ground Truth to create labeled datasets for a text classification task. The labeling job uses a private workforce of 10 annotators. After labeling 10,000 items, the quality of labels is inconsistent. Which approach will MOST effectively improve labeling consistency?

Accepted Answer

Configure the labeling job to use annotation consolidation with majority voting and require multiple annotations per item.. Requiring multiple annotations per item and using annotation consolidation with majority voting directly improves labeling consistency by reducing individual annotator bias and ensuring that the final label is based on consensus. Option A is incorrect because removing annotations from low-agreement annotators after the job completes does not improve the quality of labels already assigned; it may also discard valid data. Option B is incorrect because simply increasing the number of annotators does not guarantee consistency; it may introduce more variance without a consolidation mechanism. Option D is incorrect because active learning is used to select which items to label, not to improve the consistency of labeling; it does not address the inconsistency in the labeling process itself.

Answer

Remove annotations from annotators with low agreement after the job completes.

Answer

Increase the number of annotators to 20 to average out inconsistencies.

Answer

Use active learning to automatically label the most confident samples and only send uncertain ones to annotators.

Question 11

A research team is training a deep learning model for object detection using SageMaker's built-in SSD algorithm. The dataset contains 50,000 images with bounding box annotations. The team uses a single ml.p3.2xlarge instance. After 24 hours of training, the model's loss has plateaued, but the mean average precision (mAP) on validation is only 0.45. The team wants to improve mAP without increasing training time. Which action should they take?

Accepted Answer

Use a pre-trained model as the backbone (e.g., ResNet-50 pre-trained on ImageNet). Using a pre-trained backbone such as ResNet-50 trained on ImageNet gives the SSD model a strong feature extractor from the start, which typically improves mAP significantly without increasing training time. Transfer learning leverages general visual features (edges, textures, shapes) learned on a large dataset, allowing the model to converge to better detection performance on the 50,000-image dataset.

Answer

Increase the learning rate by a factor of 2

Answer

Increase the batch size to 64

Answer

Add more convolutional layers to the backbone

Question 12

A data scientist is using SageMaker to train a model using the built-in XGBoost algorithm. The training job fails with the error 'AlgorithmError: Framework error: No module named 'xgboost''. What is the most likely cause?

Accepted Answer

The training job is using a custom container that does not have XGBoost installed.. The error 'No module named xgboost' indicates the Python environment inside the training container cannot import the XGBoost library. When using SageMaker's built-in XGBoost algorithm, the correct algorithm container already includes XGBoost; if the job is pointed at a custom container (or a generic framework container) without XGBoost installed, this import error occurs. The most likely cause is that the training job is using a custom container lacking XGBoost.

Answer

The training data is not in CSV format.

Answer

The IAM role does not have permission to access SageMaker.

Answer

The S3 output path is incorrect.

Question 13

A data scientist needs to create a SageMaker notebook instance with access to a private S3 bucket. The bucket uses SSE-KMS encryption. Which additional configuration is required?

Accepted Answer

Attach a policy to the notebook's IAM role that allows kms:Decrypt. When an S3 bucket uses SSE-KMS encryption, the SageMaker notebook instance's IAM role must include a policy that allows the kms:Decrypt action. This is necessary because SageMaker needs to decrypt the data using the KMS key when reading objects from the bucket. Without this permission, the notebook instance will fail to access the encrypted S3 objects, even if the bucket policy allows s3:GetObject.

Answer

Add a lifecycle configuration script

Answer

Modify the bucket policy to allow s3:GetObject

Answer

Place the notebook instance in a VPC

Question 14

A data scientist is training a linear regression model and observes that the training loss is low but validation loss is high. Which step should the data scientist take to address this issue?

Accepted Answer

Apply L2 regularization to the model. Low training loss with high validation loss is the classic signature of overfitting. L2 regularization (ridge) adds a penalty term proportional to the square of the weights to the loss function, which constrains model complexity and reduces variance, thereby improving validation performance. This is the standard first-line remedy for overfitting in linear regression.

Answer

Increase the number of training epochs

Answer

Reduce the size of the training dataset

Answer

Add more features to the model

Question 15

Refer to the exhibit. A data scientist checks the status of a SageMaker endpoint and sees the output above. The endpoint is receiving traffic, but the data scientist notices that the number of instances has not increased to the desired count. What is the most likely reason?

Accepted Answer

Automatic scaling is not configured for the endpoint. The endpoint is receiving traffic but not scaling out, which indicates that automatic scaling (Application Auto Scaling) has not been configured for the SageMaker endpoint. Without a scaling policy, the endpoint will only use the initial instance count, regardless of traffic load. The status shown does not indicate any update or quota issue, so the lack of scaling is the most likely cause.

Answer

The endpoint is performing a rolling update

Answer

The endpoint is currently being updated

Answer

The account has reached its instance limit

Question 16

A data science team is using Amazon SageMaker to train a model. The training job is failing with an 'OutOfMemory' error. The team is using a p3.2xlarge instance with 61 GB of memory. They need to resolve this issue as quickly as possible. Which action should they take?

Accepted Answer

Use a larger instance type, such as p3.8xlarge. The 'OutOfMemory' error indicates the training job requires more memory than the 61 GB available on the p3.2xlarge instance. The fastest resolution is to scale vertically by using a larger instance type, such as the p3.8xlarge, which provides 244 GB of memory. This directly addresses the memory shortage without requiring code changes or architectural modifications, minimizing downtime.

Answer

Reduce the batch size in the training script

Answer

Use a spot instance to save costs

Answer

Enable distributed training across multiple instances

Question 17

A data scientist is using Amazon SageMaker to train a classification model. The dataset contains categorical features with high cardinality. Which encoding method is most appropriate for handling high-cardinality categorical features in a linear model?

Accepted Answer

Target encoding. One-hot encoding creates many binary columns, which can cause the curse of dimensionality for high-cardinality features. Label encoding assigns arbitrary integers, which linear models may interpret as ordinal. Target encoding (mean encoding) replaces categories with the mean of the target variable, which captures information without expanding dimensionality. This is often used for high-cardinality features. Ordinal encoding is similar to label encoding.

Answer

Label encoding

Answer

One-hot encoding

Answer

Ordinal encoding

Question 18

A machine learning engineer is using Amazon SageMaker to train a deep learning model. The training job is taking longer than expected. The engineer notices that the GPU utilization is low (around 30%) while CPU utilization is high. Which action is most likely to improve training speed?

Accepted Answer

Increase the number of data loading workers. Low GPU utilization with high CPU utilization indicates the data pipeline (host-side preprocessing, augmentation, disk I/O) cannot feed the GPU fast enough. Increasing the number of data loading workers parallelizes data fetching and preprocessing, raising GPU utilization and reducing training time.

Answer

Use a smaller instance type with fewer GPUs

Answer

Decrease the number of data loading workers

Answer

Increase the batch size

Question 19

Which TWO actions are appropriate when handling missing data in a dataset for machine learning? (Select TWO.)

Accepted Answer

Use a machine learning model to predict missing values based on other features. Option A is correct because using a machine learning model to predict missing values based on other features (a technique known as model-based or regression imputation, e.g., k-NN or MICE) leverages correlations between features to produce more accurate estimates than simple statistics, making it an appropriate handling strategy. Option C is correct because imputing missing values with the mean or median of the feature is a standard, simple statistical imputation method that preserves the dataset size and is especially suitable for numerical features, with the median being robust to outliers. Option B is not appropriate because dropping all rows with any missing value can discard a large portion of the data and introduce bias, especially if missingness is not random. Option D is not appropriate because removing an entire feature with missing values discards potentially useful predictive information that could otherwise be recovered through imputation. Option E is not appropriate because filling missing values with zero is an arbitrary constant that can distort the feature's distribution and introduce false signal unless zero is a meaningful domain value.

Answer

Drop all rows that contain any missing value

Answer

Remove the feature entirely if it contains missing values

Answer

Fill missing values with zero

Question 20

A data scientist is training a linear regression model and notices high bias in the training set. What action is most likely to reduce bias?

Accepted Answer

Add more relevant features to the model.. High bias indicates that the model is underfitting the training data, meaning it is too simple to capture the underlying patterns. Adding more relevant features increases the model's capacity to learn complex relationships, directly reducing bias. This is a standard approach in linear regression to address underfitting.

Answer

Apply L1 regularization.

Answer

Increase the learning rate.

Answer

Increase the amount of training data.