AWS Certified Machine Learning Engineer - Associate MLA-C01

Sign up for an AWS free tier account, explore the three free offers—free trials, 12 months, and always free—and learn budgeting and identity verification steps for a hands-on start.

Explore SageMaker as an end-to-end machine learning lifecycle solution, covering data preparation, building and training models, deployment to endpoints, and a unified studio interface with notebooks.

Explore SageMaker notebooks, managed Jupyter environments for data exploration, model training, and deployment without managing infrastructure. Compare notebook instances and studio notebooks, with persistent storage and TensorFlow and PyTorch.

Sign in to the AWS console and set up SageMaker notebooks in a managed Jupyter environment for interactive coding, data exploration, and visualizations, including creating a notebook instance with ml.t2.medium.

Explore basic SageMaker notebook operations by creating a PyTorch notebook, loading image data, training a model to recognize handwritten digits, and managing the instance to control costs.

Learn how to set up a SageMaker domain for your organization, create user profiles with private repositories and EFS storage, and enable access to SageMaker Studio and related apps.

Explore SageMaker Studio interface, including the home menu and navigation, and learn to set up a Jupyter Lab space while exploring Data Wrangler, Feature Store, AutoML, experiments, and pipelines.

Set up an AWS budget to stay within the free tier and receive email alerts when costs exceed thresholds, while exploring billing, cost management, and Cost Explorer tools.

Explore SageMaker data Wrangler to simplify data preparation for machine learning, cleaning and transforming data, applying one-hot encoding and normalization, and exporting ready data to SageMaker or S3.

Explore data wrangler in SageMaker Studio and Canvas to import, clean, and visualize data without code, then build a data flow to prepare and export for model training.

Explore the data wrangler data flow, visualize data with histograms and correlations, identify quality issues, generate insights, and create analyses like bias report and scatter plot to inform feature selection.

Add transformation steps to prepare data for training, including handling missing values and addressing outliers. Impute missing values, replace incorrect entries, encode categories, drop columns, and export the cleaned data.

Export and reuse data flows from data wrangler by exporting to canvas or S3, or as a reusable Jupyter notebook, with pre-export validation and optional data joins.

Stop unused resources by stopping running instances to prevent unnecessary charges, and export features like weather data and holidays to a feature store for reuse across models.

Explore how feature engineering transforms raw data into meaningful features—handling missing values, scaling, encoding categoricals, creating interaction terms, and selecting features to boost model performance for SageMaker feature store.

Utilize the SageMaker feature store as a centralized repository to store, manage, and reuse features, including feature groups, online (low latency) and offline stores for real-time and batch ML pipelines.

Explore the SageMaker Studio feature store to create, ingest, and share features and feature groups, including online and offline storage, data ingestion, and Athena and Glue queries.

Set up a Jupyter Lab space, upload and run a notebook to create feature groups and ingest data into the SageMaker feature store, then verify with Getrecord and Athena.

Explore SageMaker Ground Truth to label data with automated workflows, using active learning, including object detection, image and text classification, 3D point cloud labeling, and human in the loop.

Discover how SageMaker ground truth creates labeled data for supervised training, using image and text classifications, and how to build private labeling workflows with an email-invited team.

Create a labeling job in Ground Truth, configure input and output in S3, prepare the manifest, assign an execution role, and generate output for image classification.

Leverage Ground Truth Plus to access a fully managed data labeling workflow, with setup, worker management, quality control, and output, plus project requests, expert guidance, and progress tracking.

Explore SageMaker built-in algorithms, including XGBoost and linear learner, and use custom algorithms with TensorFlow or PyTorch, training with pre-packaged containers by supplying data and hyperparameters.

Explore SageMaker Jumpstart to rapidly deploy ML models with foundation and pre-trained models, including text and image generation, via SageMaker Studio or the Python SDK, with prompt engineering and fine-tuning.

Explore SageMaker Jumpstart to deploy pre-trained models for image classification, text generation, and more, with one-click endpoint deployment and notebook-based inference.

Explore three main training paths: no-code/low-code in canvas, code-based training with notebooks, and custom containers, using built-in algorithms, Jumpstart models, or your own data.

Set up a Jupyter Lab space, upload employee dataset, and prepare data for an XGBoost binary classification model in SageMaker, transforming categorical features to numeric and configuring the feature store.

Save and upload training and validation data to S3, configure a SageMaker estimator with hyperparameters, and train using a CSV input. Track RMSE and auto billing during training.

Review the trained model's evaluation and artifacts, including the root mean square error of 0.33 and training/validation data artifacts, then prepare for hyperparameter tuning to improve performance.

Explore how hyperparameters, external configurations guiding learning, differ from model parameters and influence generalization. See how SageMaker automatic tuning runs training jobs across hyperparameter ranges to optimize a chosen metric.

Explore hyperparameter tuning techniques for machine learning models, including grid search, random search, bayesian optimization, and Hyperband, with emphasis on efficiency, exploration, and early stopping of underperforming configurations.

Tune hyperparameters in notebooks using SageMaker's hyperparameter tuner to optimize XGBoost models, compare RMSE, and retrain with the best configuration.

Explore hyperparameter tuning in the SageMaker UI by creating tuning jobs, using warm start and early stopping, and choosing random or Bayesian search strategies for XGBoost training.

Explore SageMaker canvas, a no-code platform for end-to-end machine learning and generative AI, featuring AutoML autopilot, data wrangler, data flows, and one-click deployment.

Evaluate model performance in SageMaker Canvas, explore predictions, batch and single, and deploy to a real-time endpoint for production, with explainability insights and cost-conscious stopping.

Leverage SageMaker script mode to run custom models in pre-built framework containers, using entry point scripts and optional dependencies. Train with an estimator and deploy for real-time or batch inference.

Create custom docker containers for SageMaker by packaging code, dependencies, and the operating system into a single image, then push to ECR for distributed training and real-time or batch inference.

Scale training across multiple machines with distributed training and model parallelism in SageMaker. Learn data parallelism, PyTorch model parallelism, and how SageMaker handles data distribution and gradient synchronization.

Explore SageMaker experiments to organize multiple training runs with varied hyperparameters and algorithms, capture metrics, and identify the best configuration using MLflow integration.

Set up and use MLflow tracking server to log experiments, compare runs by hyperparameters and metrics, track data features and artifacts, and manage models for SageMaker.

Track, log, and compare ML experiments with MLflow in SageMaker—set up experiments and runs, log hyperparameters, metrics, and artifacts, compare multiple runs, and register models.

Delete the MLflow tracking server to avoid storage costs from the EBS volume, by navigating to MLflow, selecting the server, choosing delete, confirming, and waiting for deletion.

Explore responsible ai across traditional and generative models, emphasizing trustworthy, transparent systems throughout design, development, deployment, monitoring, and evaluation, while addressing bias, variance, and the bias-variance tradeoff.

Apply strategies to reduce bias and variance in machine learning models, including cross validation, regularization, dimensionality reduction, and early stopping for better generalization and responsible AI.

Use SageMaker Clarify to perform pre-training bias analysis on the training data and post-training bias and explainability analyses, generating reports that reveal gender-based biases and feature contributions.

Explore the pre-training analysis results from SageMaker Clarify, examining bias metrics, class imbalance, and label distribution to assess training data bias before post-training analysis.

Train and deploy a model, then run SageMaker Clarify post-training bias analysis to compare pre- and post-training bias metrics, including accuracy, acceptance rate, and explainability insights.

Explains how shap values power the SageMaker Clarify explainability report, yielding global and local model explanations from individual samples, with outputs stored in S3 and a downloadable visualization report.

SageMaker debugger, integrated in SageMaker, enables real-time monitoring and debugging during training, with alarms, metrics visualization, and support for TensorFlow, PyTorch, and XGBoost.

Demonstrate how the SageMaker debugger monitors training, detects overfitting and loss stalls, and uses debugger hooks and trials to improve model accuracy from 74% to 83%.

Profile training with sageMaker to monitor compute resources. Identify bottlenecks like gpu underutilization or slow disk reads and optimize training with real-time dashboards.

Explore SageMaker model deployment strategies by comparing real-time, serverless, batch transform, and asynchronous endpoints, and learn when to use each to optimize performance and cost in production.

Explore deploying models through real-time endpoints, including serverless, batch transform, asynchronous inference, and multi-model endpoints. Prepare data and train with xgboost, then deploy and test fast real-time predictions.

Deploy a real-time endpoint using a model artifact from a pre-trained model. Use the SageMaker runtime to invoke the endpoint and generate predictions, then delete the endpoint when finished.

deploy a serverless inference endpoint to deploy models without managing infrastructure, using memory size and maximum concurrency to scale automatically and run predictions cost-effectively for varying traffic.

Deploy using batch transform to run inference on large data sets stored in S3, outputting predictions to a CSV file and managing resources on demand for cost efficiency.

Deploy an asynchronous inference endpoint to handle long-running requests and store results in an s3 location. Submit data, queue processing, and configure the async output path and max concurrent invocations.

Use SageMaker multi-model endpoints to host multiple models on one endpoint, saving costs and reducing overhead, with on-demand loading from S3, and memory caching.

Learn to deploy a multi-model endpoint in SageMaker, loading models on demand to run real-time inferences from a single endpoint while reducing costs.

SageMaker Neo optimizes models for performance across cloud and edge devices by compiling for target hardware while preserving accuracy.

Learn how to monitor models in production, track data drift, model quality drift, bias drift, and feature attribution drift, and trigger retraining or updates using SageMaker model monitor.

Explore Amazon SageMaker model monitor to validate production models, tracking data quality, bias, explainability, and drift, with CloudWatch alerts and outputs to S3, real-time endpoints, batch transforms, and scheduled monitoring.

Monitor data quality in SageMaker with model monitor to detect data drift, capture endpoint data, alert on issues, and rely on baselines and CloudWatch.

Monitor model quality with SageMaker model monitor by merging endpoint predictions with ground truth, evaluating against a baseline, and generating CloudWatch alerts based on deviations using problem-specific metrics.

Set up a model monitor on an existing endpoint to track model quality, data quality, and bias drift, generating baselines from training data and validating inference data against constraints.

Create a SageMaker model monitor schedule to assess data quality against baseline constraints, generate hourly processing reports, and handle violations while managing endpoints.

Explore SageMaker pipelines, a serverless, scalable workflow automation tool for MLOps that uses a drag-and-drop UI or pipelines SDK to define DAGs, steps, and reusable ML workflows in production.

Learn to build and run SageMaker pipelines in both the Jupyter notebook and the SageMaker Studio visual interface, creating repeatable training to model registration workflows.

Explore SageMaker model registry to centralize and manage model packages, versions, and metadata, enabling automated registration and easy deployment to real-time endpoints or batch transforms via pipelines.

Learn to register and catalog models in the SageMaker model registry, create model groups, manage versions, set approval status, and attach training metrics and metadata.

Delete the SageMaker domain and all user profiles to remove setup resources. Also delete the corresponding file system (EFS) to complete resource cleanup and prevent extra costs.

Explore supervised, unsupervised, and reinforcement learning, and see how classification, regression, and clustering tackle real problems like spam detection, image classification, and house price prediction using neural networks or XGBoost.

Explore supervised learning with labeled data, training models like linear learner, XGBoost, deep R for time series, factorization machines, KNN, and embeddings to predict outcomes and enable recommendations.

Explore unsupervised learning, where the algorithm finds patterns in unlabeled data. PCA reduces features, k-means clusters data, and anomaly detection with random cut forest flags outliers.

Discover SageMaker text analysis algorithms, including BlazingText for fast classification, word embeddings, sequence-to-sequence tasks, and topic modeling with LDA and neural topic models, with outputs like topics and sentiment.

Explore image processing in SageMaker: classify images, perform semantic segmentation, and detect objects with bounding boxes. Learn data preparation, algorithm selection, training, and evaluation for practical AI tasks.

Explore reinforcement learning basics, where an agent interacts with an environment to maximize rewards through actions and states, using policy gradient and reward functions, including SageMaker reinforcement learning.

Explore how SageMaker reinforcement learning lets you build, train, and deploy RL models at scale with managed infrastructure, pre-built algorithms, a reinforcement learning toolkit, and flexible environments.

Evaluate model performance using a dedicated evaluation data set split into training, validation, and test sets to assess accuracy and predictive quality.

Explore classification and regression evaluation metrics, including accuracy, precision, recall, F1, and thresholds, plus the area under the receiver operating characteristic curve, MAE, MSE, RMSE, and R-squared for model performance.

Explore the end-to-end machine learning development lifecycle, from identifying business goals and problem framing to data collection, pre-processing, feature engineering, and deploying and monitoring models.

Explore MLOps as the extension of DevOps for machine learning, aligning development, deployment, monitoring, and retraining with automation, CI/CD, governance, and collaboration across teams using SageMaker tools.