Docs suggest role is optional in local mode

[stevehawley]

Please fill out the form below.

System Information

• Framework (e.g. TensorFlow) / Algorithm (e.g. KMeans):
  TensorFlow

• Framework Version:
  1.12

• Python Version:
  3.7

• CPU or GPU:
  NA

• Python SDK Version:

• Are you using a custom image:
  No

Describe the problem

Documentation at https://sagemaker.readthedocs.io/en/stable/overview.html#local-mode shows creating estimator without passing role, when doing this locally I get an error that role is a required positional parameter.

specifically this code:

from sagemaker.mxnet import MXNet

mxnet_estimator = MXNet('train.py',
                        train_instance_type='local',
                        train_instance_count=1,
                        framework_version='1.2.1')

mxnet_estimator.fit('file:///tmp/my_training_data')

And why would an IAM role be required for local mode anyway?

Minimal repro / logs

from sagemaker.tensorflow import TensorFlow

mnist_estimator = TensorFlow(entry_point='mnist.py',
                             training_steps=10, 
                             evaluation_steps=10,
                             train_instance_count=1,
                             train_instance_type='local')

logs:

Traceback (most recent call last):
  File "sm_driver.py", line 8, in <module>
    train_instance_type='local')
  File "/home/sdhawley/work/cohereum/mnist/.venv/lib/python3.7/site-packages/sagemaker/tensorflow/estimator.py", line 235, in __init__
    super(TensorFlow, self).__init__(image_name=image_name, **kwargs)
  File "/home/sdhawley/work/cohereum/mnist/.venv/lib/python3.7/site-packages/sagemaker/estimator.py", line 805, in __init__
    super(Framework, self).__init__(**kwargs)
TypeError: __init__() missing 1 required positional argument: 'role'

• Exact command to reproduce:

[mvsusp]

Hi @stevehawley

Thanks for your observation. Although role is not necessary in local mode, it allows a smoother transition to other SageMaker services, which is why we keep the parameter.

I updated the documentation as suggested.

Thanks for using SageMaker

[zhijinl]

Sorry to bring this up again. @mvsusp Thanks for giving an answer to it, but I still don't get it: "Although role is not necessary in local mode ...", so we do agree that "role" needs to be an optional parameter, not a required one right?

Making it "required" means that you need to have an AWS account, which many people don't, and won't prefer to be forced to have one.

[janhurst]

Making it "required" means that you need to have an AWS account, which many people don't, and won't prefer to be forced to have one.

This is a pain point for me too... it seems local mode isn't really local.

I have to specify a role and that role is being validated before I can run the training job locally. I was expecting this could be done completely separately from AWS without needing any credentials.

Unfortunately, for my use case, I can't have an AWS account and a role being defined as a dependency.