Document support of inference_response_keys on AutoML.deploy()

[athewsey]

What did you find confusing? Please describe.

AutoML.create_model() supports an inference_response_keys parameter which controls what kind of results the created model produces: e.g. Label, top label confidence and/or all confidences for multiclass classification.

AutoML.deploy() doc does not mention the inference_response_keys parameter or any default behaviour of additional kwargs...

...However, it is possible to pass inference_response_keys in and the created endpoint will respond as specified.

Describe how documentation can be improved

Best would be to explicitly support this argument, because it's important for how the results of the model will be consumed as it significantly changes the API contract.

Documentation could also be improved to indicate what parent/downstream methods unlisted kwargs will be sent to.

Additional context

N/A

[chuyang-deng]

Hi @athewsey , I think AutoML.deploy() does have a docstring regarding inference_response_keys, is this what you are looking for? https://github.com/aws/sagemaker-python-sdk/blob/master/src/sagemaker/automl/automl.py#L397

[athewsey]

Thanks @ChuyangDeng - Agreed this is now present on the AutoML.deploy() doc as I originally linked: I assume the stable tag just updated or something since I saw it!

Docs look fine on this feature, so happy to close.