Can't define OutputDataConfig for TrainingStep #80
Comments
|
+1 on this. One could potentially work around this by overwriting the sagemaker default bucket (aws/sagemaker-python-sdk#1402), but this doesn't allow you to specify a path, only a different bucket. As the API integration supports creating the OutputDataConfig (https://docs.aws.amazon.com/sagemaker/latest/APIReference/API_CreateTrainingJob.html), it makes sense to expose it via this SDK directly. |
shivlaks
pushed a commit
that referenced
this issue
Jun 19, 2021
…n for InputDataConfig and OutputDataConfig (#142) Currently, it is not possible to specify the output path to TrainingStep. It needs to be defined at the Estimator level, which does not support placeholders. This change will make it possible to insert a placeholder output path from the TrainingStep definition and propagate it dynamically to the Estimator. This change change also makes the TrainingStep data parameter compatible with placeholders. There are other feature requests to make other TrainingStep arguments and ProcessingStep arguments compatible with placeholders. They will be addressed in a separate PR where the implementation could maybe be done at a higher level to avoid repetition. Closes #98 #97 #80
|
Implemented in #142 which has now been merged 🎉 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
It doesn't look like it's possible to specify the output path for the model artifacts in the TrainingStep class. You can only use the default output directory or define it in the estimator object before you pass it to the TrainingStep. In either case it doesn't look like its possible to set the output path from the execution input or the output of a prior task.
The text was updated successfully, but these errors were encountered: