Features

• Support placeholders for:
  • TransformStep (#157)
  • Processing Step (#155)
  • Tuning Step (#173)
• Add custom description for CloudFormation template (#161)
• Support for following service integrations:
  • AWS Step Functions service integration (#166)
  • AWS Glue Databrew service integration (#176)
  • Amazon EKS service integration (#177)

Fixes

• Placeholder hyperparameters passed to TrainingStep are overwritten (#152 )
• Environment variables are overwritten and not passed through to SageMaker ModelStep (#160)
• Environment variables defined in the Estimator are not translated to Model when calling TrainingStep.get_expected_model (#160)
• Retrier and Catcher passed to constructor for Task, Parallel and Map states are not added to the state's Retriers and Catchers (#169)

Features

• Placeholders support in TrainingStep to set S3 location for InputDataConfig and OutputDataConfig (#142)
• EventBridge service integration (#147)

Fixes

• supplying hyperparameters to training step constructor drops hyperparameters specified in estimator (#144)

Features

• Make Choice states chainable#132

Bug fixes

• Make service integration ARNs partition aware#131

⚠ BREAKING CHANGES

Python Support

• Drops support for Python 2

SageMaker SDK

• If your project is using the Amazon SageMaker Python SDK, it must be upgraded toversion 2.x. Read more about upgrading toversion 2.x.

StepFunctions SDK

• TrainingStep -sagemaker.session.s3_input has been renamed tosagemaker.inputs.TrainingInput
• TuningStep -sagemaker.session.s3_input has been renamed tosagemaker.inputs.TrainingInput
• TrainingPipeline -sagemaker.session.s3_input has been renamed tosagemaker.inputs.TrainingInput
• InferencePipeline -sagemaker.session.s3_input has been renamed tosagemaker.inputs.TrainingInput

Features

• New choice state comparators that were added to theAmazon States Language specifications (#88) (08a0ed7)
• Add support for dynamic timeout and heartbeat in Task state (#113) (7640db1)

• Updgrading to SageMaker SDK v2
• Sunsetting support for Python 2

• Support booleanFalse fordebugger_hook_config parameter inTrainingStep
• Expose TuningStep like other SageMaker steps under thesteps module

Bug fixes and small improvements.