AWS SageMaker for Data Scientists Cheat Sheet

Updated 2026-05-15

Next Topic: Bayesian Statistics with PyMC Cheat Sheet

🧠Study flashcards on this topic119 cards · spaced repetition→

Amazon SageMaker is a fully managed machine learning service that enables data scientists to build, train, and deploy ML models at scale within a unified, production-ready environment. As part of AWS's broader AI platform, SageMaker provides an end-to-end workflow from data preparation through model deployment, removing the infrastructure overhead that traditionally slows down ML development. What makes SageMaker particularly valuable is its serverless orchestration — you define pipelines and training jobs, and AWS automatically provisions, scales, and shuts down compute resources as workload demands change, letting you focus entirely on optimizing model performance rather than managing clusters.

What This Cheat Sheet Covers

This topic spans 22 focused tables and 165 indexed concepts, 119 flashcards. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: SageMaker Studio IDE ComponentsTable 2: Built-in Algorithms for Tabular DataTable 3: Training Job ConfigurationTable 4: SageMaker Pipelines for ML OrchestrationTable 5: SageMaker Feature StoreTable 6: SageMaker Model MonitorTable 7: SageMaker Clarify for Bias and ExplainabilityTable 8: Real-Time Inference EndpointsTable 9: Data Wrangler for No-Code Data PreparationTable 10: SageMaker Experiments for Run TrackingTable 11: SageMaker Canvas for No-Code AutoMLTable 12: Distributed Training StrategiesTable 13: Hyperparameter TuningTable 14: Batch Transform for Offline InferenceTable 15: Processing Jobs for Data/Model TasksTable 16: Model Registry and DeploymentTable 17: Asynchronous InferenceTable 18: SageMaker Neo for Model OptimizationTable 19: SageMaker JumpStartTable 20: Spot Training for Cost OptimizationTable 21: SageMaker Autopilot for AutoMLTable 22: Advanced Training Features

Quick IndexSubscribe to unlock

A jump-to index of every table row in this cheat sheet.

Mind MapSubscribe to unlock

An interactive map of every table and concept in this topic.

Table 1: SageMaker Studio IDE Components

Studio is the web-based workbench where most SageMaker work begins, and these components define how teams and their compute are organized inside it. A Domain draws the team boundary, user profiles isolate each person's workspace, and the kernel gateway plus instance type determine what hardware actually runs your notebook cells — pick a GPU image here and your interactive experiments get real acceleration.

Component	Example	Description
SageMaker Domain	`domain = session.create_domain(` `domain_name="ml-team"` `)`	• Managed environment that provides authentication, authorization, and resource isolation for teams • acts as organizational boundary containing user profiles and shared storage
User Profile	`profile = domain.create_user_profile(` `user_profile_name="data-scientist-1"` `)`	• Individual workspace within a domain with dedicated storage and IAM execution role • each user gets isolated Jupyter environment
JupyterLab Application	Launch JupyterLab 4 from Studio UI	• Interactive notebook environment with support for Python, R, and custom kernels • provides code editing, debugging, and Git integration
Kernel Gateway	Select `Python 3 (Data Science 3.0)` kernel	• Compute backend that runs notebook cells • supports multiple kernel images with pre-installed ML frameworks
SageMaker Image	`Data Science 3.0`, `PyTorch 2.0`, `TensorFlow 2.13`	• Pre-built container images with ML frameworks and libraries • can bring custom Docker images for specific dependencies
Instance Type	`ml.t3.medium` (2 vCPU, 4 GB) `ml.g4dn.xlarge` (4 vCPU, 16 GB, 1 GPU)	• Compute instance powering the kernel gateway • CPU instances for lightweight work, GPU instances for training/inference
Lifecycle Configuration	Shell script to install packages on start	• Automation scripts that run on kernel start or notebook creation • useful for installing custom dependencies
Shared Storage	`/home/sagemaker-user/` directory	• Amazon EFS-backed persistent storage shared across all notebooks in user profile • survives kernel shutdowns

Table 2: Built-in Algorithms for Tabular Data

For the structured, columnar data that dominates real business problems, SageMaker ships managed implementations you can train without writing the model yourself. The gradient-boosting trio — XGBoost, LightGBM, and CatBoost — is what most practitioners reach for first, while AutoGluon-Tabular ensembles them automatically and TabTransformer brings deep self-attention to tables when feature interactions run deep.

Algorithm	Example	Description
XGBoost	`from sagemaker.xgboost import XGBoost` `xgb = XGBoost(..., objective="reg:squarederror")`	• Gradient-boosted decision trees for regression and classification • highly optimized for tabular data with GPU support for faster training

The rest of this section is part of the full cheat sheet

Table 3: Training Job Configuration

The Estimator is the central abstraction for every training run — call fit() and SageMaker spins up the instances, pulls your data from S3, runs your script, and tears everything down. The knobs collected here are where the real cost and reliability tradeoffs live: Spot instances cut the bill by up to 90% as long as you checkpoint, and input modes decide whether your dataset is downloaded whole or streamed.

Technique	Example	Description
Estimator Class	`from sagemaker.estimator import Estimator` `est = Estimator(image_uri="123.dkr.ecr...",` `role=role, instance_count=1)`	• Core abstraction for training jobs • wraps training script, instance configuration, and hyperparameters into single object

The rest of this section is part of the full cheat sheet

Table 4: SageMaker Pipelines for ML Orchestration

Pipelines turn a one-off notebook workflow into a repeatable, production DAG of steps — processing, training, evaluation, and registration wired together with dependencies. Condition steps add the approval gates that separate experimentation from production, and parameters let one pipeline definition serve as a reusable template across projects.

Component	Example	Description
Pipeline Definition	`from sagemaker.workflow.pipeline import Pipeline` `pipeline = Pipeline(` `name="training-pipeline",` `steps=[process, train, evaluate])`	• Workflow orchestration for ML • defines directed acyclic graph (DAG) of steps with dependencies and conditional execution

The rest of this section is part of the full cheat sheet

Table 1: SageMaker Studio IDE Components

Component	Example	Description
SageMaker Domain	`domain = session.create_domain(` `domain_name="ml-team"` `)`	• Managed environment that provides authentication, authorization, and resource isolation for teams • acts as organizational boundary containing user profiles and shared storage
User Profile	`profile = domain.create_user_profile(` `user_profile_name="data-scientist-1"` `)`	• Individual workspace within a domain with dedicated storage and IAM execution role • each user gets isolated Jupyter environment
JupyterLab Application	Launch JupyterLab 4 from Studio UI	• Interactive notebook environment with support for Python, R, and custom kernels • provides code editing, debugging, and Git integration
Kernel Gateway	Select `Python 3 (Data Science 3.0)` kernel	• Compute backend that runs notebook cells • supports multiple kernel images with pre-installed ML frameworks
SageMaker Image	`Data Science 3.0`, `PyTorch 2.0`, `TensorFlow 2.13`	• Pre-built container images with ML frameworks and libraries • can bring custom Docker images for specific dependencies
Instance Type	`ml.t3.medium` (2 vCPU, 4 GB) `ml.g4dn.xlarge` (4 vCPU, 16 GB, 1 GPU)	• Compute instance powering the kernel gateway • CPU instances for lightweight work, GPU instances for training/inference
Lifecycle Configuration	Shell script to install packages on start	• Automation scripts that run on kernel start or notebook creation • useful for installing custom dependencies
Shared Storage	`/home/sagemaker-user/` directory	• Amazon EFS-backed persistent storage shared across all notebooks in user profile • survives kernel shutdowns

Table 2: Built-in Algorithms for Tabular Data

Algorithm	Example	Description
XGBoost	`from sagemaker.xgboost import XGBoost` `xgb = XGBoost(..., objective="reg:squarederror")`	• Gradient-boosted decision trees for regression and classification • highly optimized for tabular data with GPU support for faster training

Table 3: Training Job Configuration

Technique	Example	Description
Estimator Class	`from sagemaker.estimator import Estimator` `est = Estimator(image_uri="123.dkr.ecr...",` `role=role, instance_count=1)`	• Core abstraction for training jobs • wraps training script, instance configuration, and hyperparameters into single object

Table 4: SageMaker Pipelines for ML Orchestration

Component	Example	Description
Pipeline Definition	`from sagemaker.workflow.pipeline import Pipeline` `pipeline = Pipeline(` `name="training-pipeline",` `steps=[process, train, evaluate])`	• Workflow orchestration for ML • defines directed acyclic graph (DAG) of steps with dependencies and conditional execution