Tags: #core #important #hands-on
Fully managed ML service for building, training, and deploying models at scale.
- Web-based IDE for ML development
- Jupyter notebooks, experiment tracking, debugging
- Managed Jupyter notebook instances
- Pre-configured with ML frameworks
- Linear Learner - Linear regression, classification
- XGBoost - Gradient boosted trees
- K-Means - Clustering
- PCA - Dimensionality reduction
- Factorization Machines - Recommendations
- DeepAR - Time series forecasting
- BlazingText - Text classification, Word2Vec
- Object Detection - Images
- Image Classification - ResNet CNN
- Semantic Segmentation - Pixel-level classification
📖 For detailed algorithm hyperparameters, see SageMaker Hyperparameters
📖 For transfer learning and fine-tuning, see SageMaker Training & Fine-Tuning
📖 For pre-built models and solutions, see SageMaker JumpStart
- AutoML - automatically builds, trains, tunes models
- White-box approach (provides notebooks)
- Supports tabular data (CSV)
- Data labeling service
- Human labelers + active learning
- Built-in labeling workflows
- Centralized repository for ML features
- Online + offline stores
- Feature versioning, sharing, discovery
- CI/CD for ML workflows
- Define, orchestrate, automate ML workflows
- Integrates with MLOps tools
📖 For detailed pipeline and CI/CD information, see MLOps CI/CD
- Pay for compute (training, inference)
- Pay for storage (S3, EBS)
- No charge for Studio (pay for underlying compute)
- Know which built-in algorithms for which use cases
- Understand endpoint vs batch transform tradeoffs
- Spot training requires checkpointing
- Serverless inference for intermittent/unpredictable traffic
- Training data must be in S3
- Built-in algorithms expect specific data formats (RecordIO-protobuf, CSV, JSON)
- Endpoint deployment can take 5-10 minutes
- SageMaker Hyperparameters - Algorithm hyperparameters in detail
- SageMaker Training & Fine-Tuning - Transfer learning, fine-tuning
- SageMaker JumpStart - Pre-built models and solutions
- MLOps & Deployment - Deployment strategies
- MLOps CI/CD - Pipelines, Model Registry
- MLOps Experiments - Experiments, TensorBoard
- MLOps Monitoring - Model Monitor
- Model Training & Evaluation
- AWS ML Algorithms
- SageMaker Clarify