Content domains
The 4 domains of MLA-C01, with every task statement and its objectives from the official guide. Study a whole domain, or drill a single task.
1Data Preparation for Machine Learning (ML)
28% of examStudy domain
1.1Ingest and store data23 q>
Knowledge of
- Common data formats and how they are ingested, including validated and non-validated formats (for example, Parquet, JSON, CSV, ORC, Avro, RecordIO)
- The core AWS data sources an ML workload reads from (for example, Amazon S3, Amazon EFS, Amazon FSx for NetApp ONTAP)
- AWS streaming sources for ingesting data as it arrives (for example, Amazon Kinesis, Apache Flink, Apache Kafka)
- The tradeoffs and use cases that separate the AWS storage options from each other
Skills in
- Pulling data out of storage services (for example, Amazon S3, Amazon EBS, Amazon EFS, Amazon RDS, DynamoDB) using accelerating options such as S3 Transfer Acceleration and EBS Provisioned IOPS
- Picking a data format that matches the access pattern (for example, Parquet, JSON, CSV, ORC)
- Loading data into SageMaker Data Wrangler and SageMaker Feature Store
- Joining data that lives in several sources, whether through code, AWS Glue, or Apache Spark
- Debugging ingestion and storage failures rooted in capacity or scaling limits
- Making the initial storage call based on cost, performance, and how the data is structured
1.2Transform data and perform feature engineering23 q>
Knowledge of
- Cleaning and transformation methods (for example, spotting and handling outliers, filling in missing values, combining records, removing duplicates)
- Feature engineering methods (for example, scaling and standardization, feature splitting, binning, log transforms, normalization)
- Encoding methods (for example, one-hot, binary, label encoding, tokenization)
- Tooling for exploring, visualizing, and reshaping data and features (for example, SageMaker Data Wrangler, AWS Glue, AWS Glue DataBrew)
- Services that transform data in flight (for example, AWS Lambda, Spark)
- Annotation and labeling services that produce high-quality labeled datasets
Skills in
- Reshaping data with AWS tooling (for example, AWS Glue, DataBrew, Spark on Amazon EMR, SageMaker Data Wrangler)
- Building and maintaining features with AWS tooling (for example, SageMaker Feature Store)
- Labeling and validating data with AWS services (for example, SageMaker Ground Truth, Amazon Mechanical Turk)
1.3Ensure data integrity and prepare data for modeling24 q>
Knowledge of
- Pre-training bias metrics across numeric, text, and image data (for example, class imbalance [CI], difference in proportions of labels [DPL])
- Ways to correct class imbalance in numeric, text, and image datasets (for example, synthetic data generation, resampling)
- Techniques for encrypting data
- Classifying, anonymizing, and masking data
- What compliance obligations imply for the pipeline (for example, PII, PHI, data residency)
Skills in
- Checking data quality (for example, with DataBrew and AWS Glue Data Quality)
- Finding and reducing bias in data such as selection bias or measurement bias, using tools like SageMaker Clarify
- Prepping data so predictions are less biased (for example, splitting, shuffling, and augmenting the dataset)
- Staging data so the training resource can load it efficiently (for example, Amazon EFS, Amazon FSx)
2ML Model Development
26% of examStudy domain
2.1Choose a modeling approach23 q>
Knowledge of
- What each family of ML algorithms can do and which business problems it suits
- Where AWS AI services fit a specific business problem (for example, Amazon Translate, Amazon Transcribe, Amazon Rekognition, Amazon Bedrock)
- How much interpretability matters when picking a model or algorithm
- The SageMaker AI built-in algorithms and the situations that call for each
Skills in
- Judging whether the available data and the problem's complexity make an ML solution feasible at all
- Weighing candidate models or algorithms against each other for a given problem
- Reaching for built-in algorithms, foundation models, and solution templates (for example, SageMaker JumpStart, Amazon Bedrock)
- Letting cost drive model or algorithm selection
- Choosing an AI service that already solves a common business need
2.2Train and refine models23 q>
Knowledge of
- The moving parts of a training run (for example, epoch, steps, batch size)
- Ways to cut training time (for example, early stopping, distributed training)
- What drives model size
- Approaches for improving model performance
- What regularization buys you (for example, dropout, weight decay, L1 and L2)
- Hyperparameter tuning strategies (for example, random search, Bayesian optimization)
- How individual hyperparameters move performance (for example, tree count in a tree-based model, layer count in a neural network)
- Bringing models trained outside SageMaker AI into SageMaker AI
Skills in
- Building models with SageMaker AI built-in algorithms and mainstream ML libraries
- Training with SageMaker AI script mode on supported frameworks (for example, TensorFlow, PyTorch)
- Fine-tuning a pre-trained model on your own dataset (for example, Amazon Bedrock, SageMaker JumpStart)
- Tuning hyperparameters (for example, with SageMaker AI automatic model tuning [AMT])
- Wiring in automated hyperparameter optimization
- Heading off overfitting, underfitting, and catastrophic forgetting (for example, via regularization, feature selection)
- Combining several trained models for better results (for example, ensembling, stacking, boosting)
- Shrinking a model (for example, changing data types, pruning, revisiting feature selection, compression)
- Tracking model versions so runs are repeatable and auditable (for example, SageMaker Model Registry)
2.3Analyze model performance23 q>
Knowledge of
- Evaluation techniques and metrics (for example, confusion matrix, heat maps, F1 score, accuracy, precision, recall, RMSE, ROC, AUC)
- How to establish a performance baseline
- Telling overfitting and underfitting apart
- What SageMaker Clarify metrics reveal about training data and models
- Convergence problems and what they look like
Skills in
- Picking the right evaluation metric, reading it correctly, and catching model bias
- Trading model performance off against training time and cost
- Running reproducible experiments on AWS services
- Measuring a shadow variant against the production variant
- Interpreting model output with SageMaker Clarify
- Debugging convergence with SageMaker Model Debugger
3Deployment and Orchestration of ML Workflows
22% of examStudy domain
3.1Select deployment infrastructure based on existing architecture and requirements23 q>
Knowledge of
- Deployment best practices (for example, versioning, rollback strategies)
- AWS deployment services (for example, Amazon SageMaker AI)
- Serving models in real time versus in batches
- Provisioning compute for production and test environments (for example, CPU, GPU)
- What each endpoint type requires (for example, serverless, real-time, asynchronous, batch inference)
- Choosing between provided and customized containers
- Optimizing models for edge devices (for example, SageMaker Neo)
Skills in
- Weighing performance, cost, and latency against each other
- Matching the compute environment to training and inference needs (for example, GPU or CPU specs, processor family, network bandwidth)
- Picking the deployment orchestrator (for example, Apache Airflow, SageMaker Pipelines)
- Deciding on multi-model or multi-container deployments
- Picking the deployment target (for example, SageMaker AI endpoints, Kubernetes, Amazon ECS, Amazon EKS, AWS Lambda)
- Settling on a deployment strategy (for example, real time, batch)
3.2Create and script infrastructure based on existing architecture and requirements23 q>
Knowledge of
- How on-demand resources differ from provisioned ones
- How scaling policies compare
- The tradeoffs and use cases behind IaC options (for example, AWS CloudFormation, AWS CDK)
- Containerization concepts and the AWS container services
- Using SageMaker AI endpoint auto scaling policies to hit scalability targets (for example, demand-based, time-based)
Skills in
- Applying practices that keep ML solutions maintainable, scalable, and cost-effective (for example, auto scaling SageMaker AI endpoints, adding Spot Instances on the fly, EC2 instances, Lambda behind endpoints)
- Automating compute provisioning and cross-stack communication (for example, CloudFormation, AWS CDK)
- Building and maintaining containers (for example, Amazon ECR, Amazon EKS, Amazon ECS, bring your own container [BYOC] with SageMaker AI)
- Placing SageMaker AI endpoints inside the VPC network
- Deploying and hosting models with the SageMaker AI SDK
- Choosing the metric that drives auto scaling (for example, model latency, CPU utilization, invocations per instance)
3.3Use automated orchestration tools to set up continuous integration and continuous delivery (CI/CD) pipelines23 q>
Knowledge of
- What AWS CodePipeline, AWS CodeBuild, and AWS CodeDeploy can do, and their quotas
- Hooking data ingestion into orchestration services
- Version control systems and everyday usage (for example, Git)
- CI/CD principles and where they land in an ML workflow
- Deployment strategies and rollback actions (for example, blue/green, canary, linear)
- How code repositories and pipelines fit together
Skills in
- Configuring and troubleshooting CodeBuild, CodeDeploy, and CodePipeline, stages included
- Structuring continuous deployment flows that trigger pipelines (for example, Gitflow, GitHub Flow)
- Automating orchestration with AWS services (for example, deploying models, automating model builds)
- Setting up training and inference jobs (for example, EventBridge rules, SageMaker Pipelines, CodePipeline)
- Adding automated tests to CI/CD pipelines (for example, integration, unit, end-to-end)
- Building retraining mechanisms and wiring them in
4ML Solution Monitoring, Maintenance, and Security
24% of examStudy domain
4.1Monitor model inference3 q>
Knowledge of
- Drift in ML models
- Techniques for watching data quality and model performance
- The ML lens design principles that bear on monitoring
Skills in
- Monitoring models once they are in production (for example, SageMaker Model Monitor)
- Watching workflows for anomalies or errors in data processing and inference
- Catching shifts in data distribution that degrade the model (for example, SageMaker Clarify)
- Measuring production model performance with A/B testing
4.2Monitor and optimize infrastructure and costs3 q>
Knowledge of
- The metrics that matter for ML infrastructure (for example, utilization, throughput, availability, scalability, fault tolerance)
- Observability tooling for chasing latency and performance problems (for example, AWS X-Ray, CloudWatch Lambda Insights, CloudWatch Logs Insights)
- Using AWS CloudTrail to log, monitor, and kick off retraining
- How instance types change performance (for example, memory optimized, compute optimized, general purpose, inference optimized)
- What the cost analysis tools offer (for example, AWS Cost Explorer, AWS Billing and Cost Management, AWS Trusted Advisor)
- Tracking and allocating cost (for example, resource tagging)
Skills in
- Setting up and using tools to troubleshoot and analyze resources (for example, CloudWatch Logs, CloudWatch alarms)
- Creating CloudTrail trails
- Standing up dashboards for performance metrics (for example, Amazon Quick Suite, CloudWatch dashboards)
- Monitoring infrastructure (for example, with Amazon EventBridge events)
- Rightsizing instance families and sizes (for example, SageMaker AI Inference Recommender, AWS Compute Optimizer)
- Tracking down and fixing latency and scaling problems
- Getting infrastructure ready for cost monitoring (for example, by applying a tagging strategy)
- Resolving capacity issues that touch both cost and performance (for example, provisioned concurrency, service quotas, auto scaling)
- Trimming costs and setting cost quotas with the right tools (for example, AWS Cost Explorer, AWS Trusted Advisor, AWS Budgets)
- Cutting infrastructure cost through purchasing options (for example, Spot Instances, On-Demand Instances, Reserved Instances, SageMaker AI Savings Plans)
4.3Secure AWS resources3 q>
Knowledge of
- The IAM roles, policies, and groups that gate access to AWS services (for example, IAM, bucket policies, SageMaker Role Manager)
- SageMaker AI security and compliance features
- Controls over network access to ML resources
- Security practices for CI/CD pipelines
Skills in
- Granting least privilege access to ML artifacts
- Writing IAM policies and roles for the users and applications that touch ML systems
- Monitoring, auditing, and logging ML systems to keep them secure and compliant
- Troubleshooting and debugging security problems
- Building VPCs, subnets, and security groups that isolate ML systems