Azure ML local Deployment: TypeError: 'NoneType' object is not subscriptable

Question

Azure ML local Deployment: TypeError: 'NoneType' object is not subscriptable

Bram Wiggers 1

I made a script to automatically deploy Azure Models using the python SDK. I want the Data Scientists in my team to easily have their models tested locally and deploy in the cloud afterwards. I use Model.deploy with a custom configuration which looks like this:

inference_config = InferenceConfig(source_directory=params['source_directory'], entry_script='score.py', environment=environment)

When I run the script in the cloud it is working perfectly, but when I run the script locally I get the following error:

Traceback (most recent call last): File "/deploy_script.py", line 14, in <module> test_deploy.test_service(service, folder, params['service_name'], delete_service_after_test=False) File "\testing_deployment.py", line 20, in test_service result = service.run(input_json) File "\azureml\core\webservice\local.py", line 72, in decorated return func(self, *args, **kwargs) File "\azureml\core\webservice\local.py", line 429, in run cleanup_on_failure=False, score_url=self.scoring_uri) File "\azureml\core\webservice\local.py", line 442, in scoring_uri overrideScorePath = self._container_override_config["scorePath"] TypeError: 'NoneType' object is not subscriptable

It seems like there is a failure when _container_override_config is accessed because it is None.

Does anyone have an idea why it is not working for this local configuration, but it is working for the cloud-based solution?

romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2020-11-27T12:45:52.8+00:00

@Bram Wiggers I think this could be an issue with the version of python or some of the dependencies mentioned in your entry script might be missing while deploying locally. Could you please check if there are any such dependencies that needs to be installed?
Christian 1 Reputation point

2020-11-29T20:58:02.363+00:00
I've run into the same problem (_container_override_config["scorePath"] cannot be accessed). To eliminate problems with wrong environment/wrong versions, I tried this tutorial notebook (unmodified):
https://github.com/Azure/MachineLearningNotebooks/blob/master/how-to-use-azureml/deployment/deploy-to-local/register-model-deploy-local.ipynb

This results in the same problem.

Steps to reproduce:

Copy Notebook to Azure ML Studio

Select Kernel (Python 3.6 - AzureML) and some running compute-instance. The Kernel used to create the Notebook ("Python 3.6") is not available in the default Azure ML Studio Notebook environment.

Run Notebook.

Error occurs with this statement: "local_service.run(sample_input)"
romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2020-11-30T07:02:37.157+00:00

@Christian Thanks for the details, We are able to replicate the same issue. I have raised a bug with the repo and team to check the functionality and the SDK. Here is the issue.
Konstantin Miller 6 Reputation points

2020-11-30T08:30:43.853+00:00

I've also run into the problem today while deploying my model locally. The issue didn't occur before. I hope a fix or a workaround can be deployed ASAP, as it's not currently possible to debug models locally.
Konstantin Miller 6 Reputation points

2020-11-30T08:52:20.37+00:00

The following workaround seems to work:
del service._container_override_config
prior to calling the local_service.run() method.

Alternatively, you can call your service via its URI: http://localhost:<port>/score

Your answer

romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2020-11-27T12:45:52.8+00:00

@Bram Wiggers I think this could be an issue with the version of python or some of the dependencies mentioned in your entry script might be missing while deploying locally. Could you please check if there are any such dependencies that needs to be installed?
Christian 1 Reputation point

2020-11-29T20:58:02.363+00:00

I've run into the same problem (_container_override_config["scorePath"] cannot be accessed). To eliminate problems with wrong environment/wrong versions, I tried this tutorial notebook (unmodified):
https://github.com/Azure/MachineLearningNotebooks/blob/master/how-to-use-azureml/deployment/deploy-to-local/register-model-deploy-local.ipynb

This results in the same problem.

Steps to reproduce:

Copy Notebook to Azure ML Studio

Select Kernel (Python 3.6 - AzureML) and some running compute-instance. The Kernel used to create the Notebook ("Python 3.6") is not available in the default Azure ML Studio Notebook environment.

Run Notebook.

Error occurs with this statement: "local_service.run(sample_input)"
romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2020-11-30T07:02:37.157+00:00

@Christian Thanks for the details, We are able to replicate the same issue. I have raised a bug with the repo and team to check the functionality and the SDK. Here is the issue.
Konstantin Miller 6 Reputation points

2020-11-30T08:30:43.853+00:00

I've also run into the problem today while deploying my model locally. The issue didn't occur before. I hope a fix or a workaround can be deployed ASAP, as it's not currently possible to debug models locally.
Konstantin Miller 6 Reputation points

2020-11-30T08:52:20.37+00:00

The following workaround seems to work:
del service._container_override_config
prior to calling the local_service.run() method.

Alternatively, you can call your service via its URI: http://localhost:<port>/score

Share via

Azure ML local Deployment: TypeError: 'NoneType' object is not subscriptable

Your answer