SageMaker Notebooks

Master SageMaker's notebook environments for data exploration, model development, and experimentation.

Studio Notebooks vs Notebook Instances

Instance Types

Choose the right instance type for your workload:

• ml.t3.medium: General development, small datasets, code writing. Most cost-effective for learning
• ml.m5.large/xlarge: Data preprocessing, medium datasets, feature engineering
• ml.c5.xlarge: CPU-intensive computations, data transformations
• ml.g4dn.xlarge: GPU-powered development, small model training, deep learning prototyping
• ml.p3.2xlarge: Serious deep learning development with V100 GPU

Lifecycle Configurations

Lifecycle configurations are shell scripts that run when a notebook instance is created or started:

• On-create scripts: Run once when the instance is first created - install persistent packages, configure tools
• On-start scripts: Run every time the instance starts - activate environments, start services
• Auto-shutdown: Configure scripts to automatically stop idle instances after a set period
• Custom environments: Install specific Python packages, Conda environments, or system tools

Working with Data

SageMaker notebooks access data primarily through Amazon S3:

• S3 integration: Use the SageMaker Python SDK or boto3 to read and write data to S3 buckets
• Local storage: Notebook instances have attached EBS volumes for temporary local storage
• EFS mounting: Mount Amazon EFS for shared, persistent file storage across notebooks
• Athena queries: Query data lakes directly from notebooks using Amazon Athena
• Data Wrangler: Use SageMaker Data Wrangler for visual data exploration and transformation

Pre-installed Kernels and Frameworks