CODEWITHSUNDEEP
pythonamazon-s3amazon-sagemaker
marcin
marcin

Reputation: 2955

SageMaker estimator `source_dir` from S3

I would like to start a training job using SageMaker TensorFlow Estimator in a script mode.
My problem is that I don't have my training code locally or in a git repo, but only in S3 "directory" and source_dir parameter requires a local file or usage of git.

Is the only way to copy the files locally from s3 (which is problematic with python) or can I do it in a nicer way?

Upvotes: 1

Views: 2114

Answers (1)

lauren
lauren

Reputation: 513

You should be able to specify an S3 path for source_dir (and an entry point path relative to your S3 "directory")

If you look at the SageMaker Python SDK code, you can see that if an S3 path is given, then the SDK doesn't look for any local files to upload to S3: https://github.com/aws/sagemaker-python-sdk/blob/master/src/sagemaker/fw_utils.py#L351-L352

Upvotes: 1

Related Questions