Free MLA-C01: AWS Certified Machine Learning Engineer - Associate Exam Questions and Answers

An ML engineer is writing a deployment script for a model that is stored in Amazon S3. The script must create a SageMaker real-time endpoint that scales based on the number of requests per instance. Put the following steps in the correct order. Step A: Register the production variant's desired instance count as a scalable target by using Application Auto Scaling. Step B: Create an endpoint configuration and an endpoint that use a production variant. Step C: Attach a target tracking scaling policy that uses the SageMakerVariantInvocationsPerInstance metric. Step D: Create a SageMaker model that references the inference container image and the model artifacts.

Work out the matching from the exhibit, then reveal.

An ML engineer is using SageMaker automatic model tuning to optimize an XGBoost model. The engineer must search the continuous range from 0.0001 to 0.5 for the eta hyperparameter. Effective values are expected to be small, and the engineer wants the tuning job to sample the range efficiently. Which configuration should the ML engineer use?

• AA. Define eta as a ContinuousParameter from 0.0001 to 0.5 with the ReverseLogarithmic scaling type.
• BB. Define eta as a ContinuousParameter from 0.0001 to 0.5 with the Linear scaling type.
• CC. Define eta as a CategoricalParameter with the values 0.0001, 0.001, 0.01, 0.1, and 0.5.
• DD. Define eta as a ContinuousParameter from 0.0001 to 0.5 with the Logarithmic scaling type.

A company wants data scientists in an IAM group to start Amazon SageMaker training jobs. The jobs must run only with an execution role named TrainingExecRole, which has access to the approved Amazon S3 bucket. Which IAM configuration will meet these requirements with the LEAST privilege?

• AA. Grant the group sagemaker:CreateTrainingJob and iam:PassRole on all roles, with a condition that limits iam:PassedToService to sagemaker.amazonaws.com.
• BB. Attach the AmazonSageMakerFullAccess managed policy to the group and add the group as a principal in the TrainingExecRole trust policy.
• CC. Grant the group sagemaker:* and iam:GetRole on the ARN of TrainingExecRole, and let SageMaker select the role when the job is created.
• DD. Grant the group sagemaker:CreateTrainingJob and iam:PassRole, with the iam:PassRole resource limited to the ARN of TrainingExecRole.

A retail company has 3 years of daily sales data for 12,000 products, along with related data such as price and promotion flags. The company needs a single model that learns patterns across the related time series and returns probabilistic forecasts for the next 28 days. Which SageMaker built-in algorithm should the ML engineer use?

• AA. DeepAR
• BB. Random Cut Forest
• CC. BlazingText
• DD. Linear Learner

A company stores training datasets in Amazon S3. Access patterns for the datasets are unpredictable, and some datasets are not read for months. The company wants to reduce storage costs automatically, without retrieval fees and without changes to application code. Which S3 storage class should the company use?

• AA. S3 Standard-IA
• BB. S3 One Zone-IA
• CC. S3 Intelligent-Tiering
• DD. S3 Glacier Flexible Retrieval

A company runs SageMaker training jobs, processing jobs, and endpoints for six ML projects in a single AWS account. Finance wants a monthly report that shows the SageMaker spend for each project. Which solution will meet this requirement with the LEAST operational overhead?

• AA. Tag SageMaker resources with a project key, activate the tag as a cost allocation tag, and group costs by that tag in AWS Cost Explorer
• BB. Enable AWS CloudTrail data events for SageMaker and group the API call counts by project in Amazon Athena queries
• CC. Use AWS Compute Optimizer to generate per-project cost reports for each SageMaker training instance type
• DD. Create an Amazon CloudWatch dashboard of SageMaker instance metrics and estimate project cost from CPU utilization

A company runs an Amazon SageMaker training job inside a VPC. The job uses a custom container image stored in a private Amazon ECR repository and reads training data from an Amazon S3 bucket. The training subnets are private and have no route to the internet. Security policy forbids any internet path, including NAT gateways. The training job fails because it cannot pull the image. Which solution will fix the failure and meet the security policy?

• AA. Create a NAT gateway in a public subnet and add a default route to it in the route table of the training subnets
• BB. Create a gateway VPC endpoint for Amazon S3 and interface VPC endpoints for the Amazon ECR API and the Amazon ECR Docker registry
• CC. Create interface VPC endpoints for the Amazon ECR API and the Amazon ECR Docker registry and keep the default route for S3 traffic
• DD. Create a gateway VPC endpoint for Amazon S3 and attach an endpoint policy that allows access to the training bucket

A bank has a labeled dataset of 400,000 loan applications in Amazon S3. About 92% of the applications are labeled approved. Compliance officers are concerned that the approval rate differs between genders in the data. No model has been trained yet, and the ML engineer must measure the imbalance before training starts. Which solution will meet these requirements?

• AA. Run a SageMaker Clarify processing job with gender as the facet and the approval column as the label, and review the pre-training bias metrics
• BB. Deploy an endpoint and create a SageMaker Model Monitor bias drift schedule that uses the dataset as its baseline
• CC. Run the training with SageMaker Debugger rules and review the tensors captured for each gender group
• DD. Train a baseline model, then run a SageMaker Clarify post-training bias analysis that compares predictions across genders

A manufacturer has inspection cameras at 40 factory sites. The sites have intermittent internet connectivity. A trained computer vision model must run on the cameras with inference latency under 50 ms, and the inference must continue to work during network outages. Which combination of steps will meet these requirements? (Select TWO.)

Select 2 answers.

• AA. Deploy the model to a SageMaker serverless inference endpoint with provisioned concurrency
• BB. Host the model on a SageMaker real-time endpoint and call it from the cameras over the internet
• CC. Run SageMaker batch transform jobs nightly and download the predictions to the cameras
• DD. Deploy the compiled model package to the cameras by using AWS IoT Greengrass
• EE. Compile the model for the camera hardware by using Amazon SageMaker Neo

An ML engineer deploys a model to an Amazon SageMaker real-time endpoint. The application requires the model to respond in under 200 ms at the 99th percentile. The ML engineer needs an Amazon CloudWatch alarm that is not affected by the time SageMaker spends on request routing and other overhead. Which metric should the alarm use?

• AA. InvocationsPerInstance with the Average statistic
• BB. ModelLatency with the p99 statistic
• CC. Invocation4XXErrors with the Sum statistic
• DD. OverheadLatency with the p99 statistic