Skip to content

AWS SageMaker — Machine Learning Platform

AWS SageMaker — Machine Learning Platform

Section titled “AWS SageMaker — Machine Learning Platform”

Amazon SageMaker is AWS’s fully managed end-to-end machine learning platform. It covers every step of the ML lifecycle — from data preparation and model training to deployment and monitoring.

In Azure terms: SageMaker ≈ Azure Machine Learning (Azure ML)

Component Description
Studio Web-based IDE for ML (notebooks, pipelines, experiments)
Notebooks Managed Jupyter notebooks with pre-built ML environments
Training Jobs Run model training on managed compute (CPU/GPU clusters)
Endpoints Deploy models as real-time HTTP endpoints
Batch Transform Run inference on large datasets (no live endpoint)
Pipelines MLOps workflow orchestration
Feature Store Centralized storage and reuse of ML features
Model Registry Version, track, and approve models before deployment
Data Wrangler Visual data preparation and transformation
Clarify Bias detection and model explainability
Experiments Track model training runs and hyperparameter comparisons
Canvas No-code ML for business analysts
Bedrock Separate service — managed access to foundation models (LLMs)
Data Prep (Data Wrangler / S3)
↓
Feature Engineering (Feature Store)
↓
Model Training (Training Job — EC2 GPU/CPU)
↓
Model Evaluation (Clarify, Experiments)
↓
Model Registry (versioning + approval)
↓
Deployment (Real-time Endpoint / Batch Transform)
↓
Monitoring (Model Monitor — drift detection)

SageMaker provisions compute, trains your model, and terminates the instance — you pay only for training time:

import sagemaker
from sagemaker.estimator import Estimator
estimator = Estimator(
image_uri='763104351884.dkr.ecr.us-east-1.amazonaws.com/pytorch-training:2.1.0-gpu-py310',
role='arn:aws:iam::123456789:role/SageMakerRole',
instance_type='ml.g4dn.xlarge', # GPU instance
instance_count=1,
hyperparameters={
'epochs': 10,
'learning-rate': 0.001,
},
output_path='s3://my-bucket/model-output/'
)
estimator.fit({'train': 's3://my-bucket/training-data/'})

Deploy trained models as REST APIs:

predictor = estimator.deploy(
initial_instance_count=1,
instance_type='ml.m5.large'
)
# Invoke the endpoint
result = predictor.predict({'input': [1.0, 2.0, 3.0]})
Terminal window
# Invoke endpoint via CLI
aws sagemaker-runtime invoke-endpoint \
--endpoint-name my-endpoint \
--content-type application/json \
--body '{"input": [1.0, 2.0, 3.0]}' \
output.json

SageMaker includes many optimized built-in algorithms (no container needed):

Algorithm Type Use Case
XGBoost Gradient boosting Classification, regression
Linear Learner Linear models Binary/multiclass classification
K-Means Clustering Customer segmentation
BlazingText NLP Text classification, word2vec
Object Detection Computer vision Detect objects in images
Semantic Segmentation Computer vision Pixel-level image labeling
DeepAR Time series Demand forecasting
Random Cut Forest Anomaly detection Fraud detection
Feature SageMaker Azure ML
IDE SageMaker Studio Azure ML Studio
Notebooks Managed Jupyter Managed Jupyter + VS Code
Training Training Jobs (managed clusters) Compute Clusters / Compute Instances
Model deployment Endpoints (real-time/batch) Online / Batch Endpoints
MLOps pipelines SageMaker Pipelines Azure ML Pipelines
Model registry SageMaker Model Registry Azure ML Model Registry
Feature store SageMaker Feature Store Azure ML Feature Store
No-code ML SageMaker Canvas Azure ML Automated ML
LLM access Amazon Bedrock Azure OpenAI Service
Experiment tracking SageMaker Experiments Azure ML Experiments / MLflow

For Large Language Models (LLMs) and foundation models, AWS uses Amazon Bedrock (not SageMaker directly):

  • Access models from Anthropic (Claude), Meta (Llama), Amazon (Titan), Mistral, Cohere
  • Serverless — no infrastructure to manage
  • Supports RAG (Retrieval-Augmented Generation) via Knowledge Bases
  • Fine-tuning and agents support

In Azure terms: Amazon Bedrock ≈ Azure OpenAI Service