Skip to content

TrainingStep's hyperparameters does not work on v2.2.0 #152

Closed
@ohbuchim

Description

@ohbuchim

I cannot set TrainingStep's 'hyperparameters' parameter using ExecutionInput on v2.2.0. v2.1.0 works well.

When I used v2.1.0 this code worked, and I was able to set hyperparameters when I executed workflow.

training_step = steps.TrainingStep(
    "SageMaker Training Step",
    estimator=sklearn,
    data={"train": sagemaker.TrainingInput(execution_input["PreprocessingOutputDataTrain"], content_type="text/csv")},
    job_name=execution_input["TrainingJobName"],
    hyperparameters=execution_input["TrainingParameters"],
    wait_for_completion=True,
)

When I used v2.2.0 I needed to add get_schema_as_dict() to ExecutionInput. After that, 'hyperparameters' was not updated by ExecutionInput. It seems that the estimator's 'hyperparameters' setting only is set (ExecutionInput is ignored).

training_step = steps.TrainingStep(
    "SageMaker Training Step",
    estimator=sklearn,
    data={"train": sagemaker.TrainingInput(execution_input["PreprocessingOutputDataTrain"], content_type="text/csv")},
    job_name=execution_input["TrainingJobName"],
    hyperparameters=execution_input["TrainingParameters"].get_schema_as_dict(),
    wait_for_completion=True,
)

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions