MLA-C01
Free AWS Certified Machine Learning Engineer - Associate Practice Test
12 exam-style questions with full explanations — no sign-up. Score yourself, then close your gaps with the full course.
Exam fee ~$1504 exam domainsLevel Intermediate2 timed practice tests in the course
✅ Free practice — no sign-up📝 Real exam-style questions💡 Detailed explanations💸 30-day money-back via Udemy
A ride-sharing company needs to ingest mobile app clickstream events at approximately 500,000 events per second. The data must be retained so that multiple independent consumer applications (fraud detection, personalization, and analytics) can each read the same stream and replay from different offsets as needed. Which service BEST meets these requirements?
Kinesis Data Streams retains records (up to 365 days with extended retention) and supports multiple independent consumers reading the same stream at different offsets via enhanced fan-out, meeting both the throughput and replay requirements.
A retail company needs to forecast weekly demand for 10,000 SKUs across 200 stores. Each SKU has related time series data such as promotions, price changes, and holiday indicators, and the business wants probabilistic forecasts (e.g., P10, P50, P90 quantiles) rather than single point estimates. Which SageMaker built-in algorithm BEST fits this requirement?
DeepAR is purpose-built for forecasting problems involving multiple related time series, natively supports categorical/dynamic features, and produces probabilistic (quantile) forecasts rather than single-point predictions.
An ML team hosts a fraud-detection model that receives sporadic traffic — sometimes no requests for hours, sometimes bursts of a few hundred requests per minute. Payloads are only a few KB, and an occasional cold-start delay of a few seconds is acceptable. The team wants to minimize cost during idle periods and avoid managing any scaling policies. Which deployment option BEST meets these requirements?
Serverless Inference automatically provisions and scales compute capacity in response to traffic, scales down to zero when idle, and charges only for the compute consumed during inference — ideal for small-payload, unpredictable, bursty workloads where a few seconds of cold-start latency is tolerable and the team doesn't want to manage scaling policies.
A company deployed a SageMaker endpoint for a fraud-detection model. Three months after launch, the fraud team reports the model is flagging far more false positives than during initial testing. The input feature statistics (means, distributions) have not changed significantly, but the relationship between features and the fraud label has shifted due to new fraud patterns. Which SageMaker Model Monitor capability specifically detects this situation?
Model quality monitoring specifically compares model predictions to actual ground truth labels (once available) to detect degradation in accuracy/precision/recall, which is exactly what's happening here — the input distribution is stable but the concept (relationship between inputs and correct output) has shifted, causing prediction quality to degrade.
A team is training a computer vision model using distributed training across 8 GPU instances that must concurrently read millions of small image files with high throughput and low latency shared file access. Which storage option should back the training job for the best performance?
FSx for Lustre is a high-performance, POSIX-compliant parallel file system designed for sub-millisecond latencies and sustained high throughput/IOPS across many compute nodes, making it ideal for distributed training on large numbers of small files.
During a SageMaker training job, training accuracy steadily rises to 99% while validation loss decreases initially but begins increasing after epoch 20, even as training loss keeps falling. Which action would MOST effectively address this issue?
The gap between rising training accuracy and worsening validation loss is the classic signature of overfitting. Adding L2 regularization penalizes large weights to improve generalization, and early stopping halts training at the point of best validation performance, directly addressing overfitting.
A media company needs to run inference on video files up to 800 MB in size, with processing times up to 45 minutes per request. Requests arrive irregularly, near-real-time responses are not required, and the company needs results automatically stored with an SNS notification sent when each job completes. Which SageMaker inference option should be used?
Asynchronous Inference natively supports payloads up to 1 GB and processing times up to 60 minutes, queues incoming requests, and has built-in SNS success/failure notifications with results written to S3 — an exact match for large, long-running, irregular inference jobs.
A financial services company runs SageMaker training jobs that take up to 8 hours and are tolerant of interruptions. The ML team wants to reduce training costs by up to 90% while ensuring that if a Spot instance is reclaimed, the job resumes from where it left off instead of restarting. Which configuration BEST meets these requirements?
Managed Spot Training in SageMaker can cut costs up to 90% versus On-Demand. Checkpointing to S3 (via the checkpoint_s3_uri parameter) allows the training job to resume from the last saved checkpoint if interrupted, and MaxWaitTimeInSeconds must be set higher than MaxRuntimeInSeconds to allow time for Spot interruptions and restarts.
A fraud detection training dataset has 2% positive fraud labels and 98% negative labels, causing the model to predict the majority class almost every time. Which action should the ML engineer take in SageMaker Data Wrangler prior to training to address this issue?
SMOTE generates synthetic samples of the minority class, rebalancing the label distribution so the model learns to distinguish both classes instead of defaulting to the majority class.
A company wants to build a domain-specific sentiment classifier for customer reviews. They have only 2,000 labeled examples and need a solution that achieves good accuracy quickly with minimal training infrastructure management. Which approach is BEST?
SageMaker JumpStart provides pre-trained foundation models that can be fine-tuned with a small labeled dataset, leveraging transfer learning to achieve strong accuracy with limited data and infrastructure effort — exactly matching the constraints described.
A company uses SageMaker Model Registry to manage model versions. They configure an EventBridge rule to automatically trigger a CodePipeline deployment pipeline whenever a model version becomes ready for production. Which specific model package event/state should the EventBridge rule filter on to invoke the pipeline?
The standard governed MLOps pattern is for a human or automated gate to approve a model version in the Registry, changing ModelApprovalStatus to 'Approved'. EventBridge captures the 'SageMaker Model Package State Change' event on this transition and triggers the downstream deployment pipeline only for approved versions.
A healthcare company's compliance policy requires that all traffic between their VPC and Amazon SageMaker (both the SageMaker API for training job management and the runtime API for real-time inference) must never traverse the public internet. Which solution satisfies this requirement?
AWS PrivateLink VPC interface endpoints for sagemaker.api and sagemaker.runtime allow all API calls for training job orchestration and real-time inference invocation to stay entirely within the AWS network, never traversing the public internet, which is required for strict compliance environments.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$109.99 $34.99 with code FREETEST33 — valid through September 7.
Get my $34.99 deal →MLA-C01 exam — quick answers
How much does the MLA-C01 exam cost?
The exam fee is approximately $150 and varies by region — confirm current pricing with the certification vendor before you book.
What topics are on the exam?
It covers 4 domains: Data Preparation for Machine Learning (ML) (28%), ML Model Development (26%), Deployment and Orchestration of ML Workflows (22%), ML Solution Monitoring, Maintenance, and Security (24%). The full course has a dedicated chapter, lab and practice-test coverage for each.
Is this practice test really free?
Yes — all questions on this page are free with explanations and no sign-up. The paid Udemy course adds two full-length timed exams, video lessons and hands-on labs.
How do I get the discount?
Use code FREETEST33 at checkout for $34.99 (list $109.99) through September 7 — the enroll button applies it automatically.
Will this prepare me for the real exam?
The questions mirror the real exam's style and are mapped to the official domains. This is exam-focused preparation — combine the free test with the full course's timed simulations to gauge your readiness.