MLS-C01 Machine Learning Implementation and Operations Practice Test 16 — 15 Questions

Question 1

A data scientist is using SageMaker to train a deep learning model. The training script uses TensorFlow and runs on a single p3.2xlarge instance. The scientist wants to reduce training time by using multiple GPUs. What should the scientist do?

Accepted Answer

Modify the training script to use Horovod for distributed training.. Modifying the training script to use Horovod (Option B) is required for distributed training across multiple GPUs. Option A (increasing instance type) only uses more GPUs on one machine if the script supports it. Option C (using Spot) does not add GPUs. Option D (changing to PyTorch) is unnecessary.

Answer

Increase the instance count to 4 without changing the script.

Answer

Switch to PyTorch framework.

Answer

Use SageMaker Managed Spot Training.

Question 2

A team has deployed a SageMaker endpoint for a sentiment analysis model. The model was trained on text data from social media. After deployment, the team notices that the model's accuracy has dropped significantly after 3 months. Which action should the team take to detect and address this issue?

Accepted Answer

Enable SageMaker Model Monitor to detect data drift and trigger a retraining pipeline.. Setting up SageMaker Model Monitor to detect drift and triggering a retraining pipeline (Option D) automates detection and correction. Option A (re-deploy) does not address root cause. Option B (CloudWatch alarm) only monitors latency, not accuracy. Option C (A/B testing) helps compare but does not detect drift automatically.

Answer

Use SageMaker A/B testing to compare with a new model.

Answer

Re-deploy the model using the same training script.

Answer

Create a CloudWatch alarm on invocation errors.

Question 3

Which TWO actions can reduce inference latency for a SageMaker real-time endpoint? (Choose 2.)

Accepted Answer

Choose a larger instance type with more compute capacity.. Using a larger instance with more CPU/memory (Option A) and compiling the model with SageMaker Neo (Option D) both reduce latency. Option B (adding instances) increases throughput but not per-request latency. Option C (asynchronous) is for different use case. Option E (batch transform) is for offline inference.

Answer

Add more instances behind the endpoint.

Answer

Use batch transform instead.

Answer

Switch to asynchronous inference.

Question 4

Which THREE are valid considerations when deploying a large deep learning model (10 GB) on a SageMaker endpoint? (Choose 3.)

Accepted Answer

Enable SageMaker Data Compression for network transfer.. Using GPU instances (Option A), enabling data compression (Option C), and using multi-model endpoints (Option D) are valid considerations. Option B (Elastic Inference) is deprecated and not recommended. Option E (serverless inference) has a payload limit and cold starts unsuitable for large models.

Answer

Use SageMaker Serverless Inference to avoid managing instances.

Answer

Attach Elastic Inference accelerators.

Question 5

Which TWO SageMaker features can be used to monitor and debug training jobs? (Choose 2.)

Accepted Answer

SageMaker Debugger. SageMaker Debugger (Option A) captures tensors and metrics during training. CloudWatch Logs (Option E) provide training script logs. Options B (Model Monitor), C (Clarify), and D (Ground Truth) are for inference, explainability, and labeling respectively.

Answer

SageMaker Model Monitor

Answer

SageMaker Ground Truth

Answer

SageMaker Clarify

Question 6

Refer to the exhibit. A developer has this IAM policy attached to an IAM role used by SageMaker. When attempting to create an endpoint, the operation fails with an access denied error. What is the MOST likely cause?

Accepted Answer

The policy is missing s3:ListBucket on the model bucket.. The policy grants permissions to create the model, endpoint config, and endpoint, but it does not include sagemaker:InvokeEndpoint (Option A). The error is likely due to missing sagemaker:InvokeEndpoint, but the question asks about creating an endpoint. Actually, creating an endpoint does not require InvokeEndpoint. The correct answer is that the policy is missing s3:ListBucket (Option B) because SageMaker needs to list objects in the bucket when accessing the model artifacts. Option C (ecr:DescribeRepositories) is not needed. Option D (sagemaker:DescribeEndpoint) is not required for creation.

Answer

The policy is missing ecr:DescribeRepositories.

Answer

The policy is missing sagemaker:DescribeEndpoint.

Answer

The policy is missing sagemaker:InvokeEndpoint.

Question 7

Refer to the exhibit. A data scientist is training a PyTorch model on a SageMaker ml.p3.2xlarge instance (16 GB GPU memory). The training fails with the shown error. Which change should the scientist make to resolve the error?

Accepted Answer

Reduce the batch size in the training script.. Reducing the batch size (Option D) reduces GPU memory usage per iteration and can resolve OOM errors. Option A (increase instance count) does not reduce per-GPU memory. Option B (use Spot) does not help. Option C (increase epochs) will still fail.

Answer

Increase the number of instances to 2.

Answer

Use SageMaker Managed Spot Training.

Answer

Increase the number of epochs.

Question 8

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 policy grants s3:* on the bucket but does not include ecr:* permissions (Option B) needed to pull the Docker image. Option A (s3:* on objects) is too broad but not the issue. Option C (iam:PassRole on role) is allowed. Option D (sagemaker:* on specific resources) is not the issue.

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 9

A data scientist is training a model using Amazon SageMaker and notices that training is taking much longer than expected. The training job uses a single ml.p3.2xlarge instance. The data is stored in S3 and is about 50 GB in size. Which action would MOST likely reduce training time?

Accepted Answer

Change the input mode from File to Pipe.. Option C is correct because using Pipe input mode streams data directly from S3 to the training algorithm without writing to disk, reducing I/O latency. Option A is wrong because increasing instance size may help but incurs higher cost and may not address data loading bottleneck. Option B is wrong because SageMaker automatically manages data channels. Option D is wrong because enabling data compression on the S3 objects does not reduce training time significantly and adds CPU overhead for decompression.

Answer

Enable automatic data sharding in the SageMaker training job.

Answer

Enable S3 server-side encryption on the training data.

Answer

Use a larger instance type, such as ml.p3.16xlarge.

Question 10

A machine learning engineer is deploying a model using Amazon SageMaker. The model is a PyTorch model that performs real-time inference with low latency requirements. The engineer wants to use automatic scaling based on the number of concurrent requests. Which SageMaker feature should be used to achieve this?

Accepted Answer

Use Amazon SageMaker automatic scaling with a target tracking scaling policy.. Option B is correct because SageMaker's application auto scaling with a target tracking scaling policy based on the SageMakerVariantInvocationsPerInstance metric automatically adjusts the number of instances. Option A is wrong because SageMaker does not have an integrated Elastic Load Balancer; it uses a built-in load balancer. Option C is wrong because SageMaker does not natively support AWS Auto Scaling groups for endpoints. Option D is wrong because Amazon API Gateway is used for REST APIs, not for scaling SageMaker endpoints.

Answer

Create an AWS Auto Scaling group for the SageMaker endpoint.

Answer

Enable Elastic Load Balancing for the endpoint.

Answer

Deploy the model behind Amazon API Gateway with a Lambda function.

Question 11

A company is using Amazon SageMaker to build a binary classification model. The dataset is highly imbalanced, with 95% negative class and 5% positive class. Which technique should be used to address the class imbalance?

Accepted Answer

Use a weighted loss function during training.. Option D is correct because using a weighted loss function during training assigns higher weight to the minority class, helping the model learn better from imbalanced data. Option A is wrong because removing the majority class reduces data size and may lose important patterns. Option B is wrong because random under-sampling can discard useful data. Option C is wrong because using accuracy as the evaluation metric is inappropriate for imbalanced data; precision/recall or AUC are better.

Answer

Use accuracy as the primary evaluation metric.

Answer

Perform random under-sampling of the majority class.

Answer

Remove all examples from the majority class.

Question 12

An ML team is deploying a model to a SageMaker endpoint for real-time inference. The model is large (2 GB) and requires GPU for low-latency inference. The team wants to minimize cost while maintaining a response time of under 200 ms. Which instance configuration and SageMaker feature would be best?

Accepted Answer

Use a GPU instance (ml.p3.2xlarge) with SageMaker Elastic Inference.. Option B is correct because using a GPU instance (ml.p3.2xlarge) with SageMaker's Elastic Inference attaches a fraction of GPU acceleration to a CPU instance, balancing cost and performance. Option A is wrong because serverless inference may not support GPU and has cold starts. Option C is wrong because multi-model endpoints are for hosting multiple models on the same instance, not primarily for latency. Option D is wrong because batch transforms are for offline inference.

Answer

Use a batch transform job on GPU instances.

Answer

Use a serverless inference endpoint with a CPU instance.

Answer

Use a multi-model endpoint on a GPU instance.

Question 13

A data scientist is training a model using Amazon SageMaker and wants to track hyperparameter tuning jobs, training jobs, and model metrics. The team also needs to compare experiments visually. Which AWS service should be used?

Accepted Answer

Amazon SageMaker Experiments. Option C is correct because SageMaker Experiments is purpose-built for tracking and comparing ML experiments. Option A is wrong because AWS Glue is for ETL, not experiment tracking. Option B is wrong because CloudWatch Logs stores logs but does not provide experiment comparison. Option D is wrong because Step Functions is for workflow orchestration.

Answer

Amazon CloudWatch Logs

Answer

AWS Glue DataBrew

Answer

AWS Step Functions

Question 14

A company is using Amazon SageMaker to train a model. The training data is stored in an S3 bucket in a different AWS account. Which IAM policy configuration is required to allow SageMaker to access the data?

Accepted Answer

Add a bucket policy allowing access from the SageMaker execution role ARN, and ensure the SageMaker execution role has an IAM policy allowing s3:GetObject on the bucket.. Option B is correct because cross-account access requires the SageMaker execution role to have an IAM policy allowing access to the S3 bucket, and the S3 bucket policy must grant access to that role. Option A is wrong because SageMaker cannot assume a role in another account without proper trust policy. Option C is wrong because the data owner's role cannot be used directly. Option D is wrong because SageMaker does not use the data owner's IAM user credentials.

Answer

Add a bucket policy that allows s3:GetObject for the SageMaker execution role's ARN.

Answer

Create an IAM user in the data owner's account and use its credentials in SageMaker.

Answer

Use the data owner's IAM role as the SageMaker execution role.

Question 15

A machine learning engineer is deploying a model on SageMaker and needs to ensure that the endpoint can handle a sudden spike in traffic. The engineer expects traffic to increase by 10x during a promotional event. Which scaling strategy should be used?

Accepted Answer

Use scheduled scaling to add instances before the event, combined with dynamic scaling for the remaining duration.. Option A is correct because scheduled scaling can add additional instances before the expected spike. Option B is wrong because dynamic scaling may not react quickly enough to a sudden 10x spike. Option C is wrong because manual scaling requires human intervention. Option D is wrong because a single large instance provides no redundancy and may not handle the spike.

Answer

Use a single large instance type instead of multiple smaller instances.

Answer

Manually increase the instance count before the event.

Answer

Use only dynamic scaling based on the average latency metric.