aws / aws/sagemaker-python-sdk

Respect default_bucket_prefix by default for Estimator code_location

Closed
#5,208 0 comments 0 reactions 0 assignees View on GitHub
component: training type: feature request
Dominant language
Python
Stars
2.3k
Forks
1.3k
Avg merge
1d 22h
Merged PRs (30d)
35

Description

**Describe the feature you'd like**
Currently the `code_location` attribute of the Estimator class defaults to using the `output_bucket` parameter [docs link](https://sagemaker.readthedocs.io/en/stable/api/training/estimators.html#sagemaker.estimator.EstimatorBase):

> If not specified, the default code location is ‘s3://output_bucket/job-name/’.

The [Session object also has a parameter](https://sagemaker.readthedocs.io/en/stable/api/utility/session.html#sagemaker.session.Session) `default_bucket_prefix` that can be configured.

Ideally, if
1. The `output_bucket` part of `output_path` is the `default_bucket`
2. `_is_output_path_set_from_default_bucket_and_prefix` is False

Then the default location should respect both the `default_bucket` as well as the `default_bucket_prefix`.

e.g. `s3://default_bucket/default_bucket_prefix/job-name/`

This change would be implemented in [_stage_user_code_in_s3](https://github.com/aws/sagemaker-python-sdk/blob/23109671f6262269ab54cdd9aeb5ebe4ea640d25/src/sagemaker/estimator.py#L1042).

Otherwise, the default behavior creates artifacts at the root of the bucket. This means that default behavior for environments where IAM bucket write access is limited by prefix (i.e. SageMaker Unified Studio) will fail.

**How would this feature be used? Please describe.**
If this behavior is implemented, model code assets would be uploaded by default to a prefix where write access is allowed.

**Describe alternatives you've considered**
Currently `code_location` needs to be manually configured to work in SageMaker Unified Studio. This is poorly documented as part of features like [ModelStep](https://sagemaker.readthedocs.io/en/stable/workflows/pipelines/sagemaker.workflow.pipelines.html#sagemaker.workflow.model_step.ModelStep) where it needs to be configured in `repack_model_step_settings` as the model.register output populates `output_path` by default in a pipeline.

If this change cannot be implemented in code, explicit documentation should be provided about configuring parameters to output code in the SageMaker Unified Studio project prefix.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.