aws-samples/ml-edge-getting-started

Machine Learning at the edge getting started

Here you can find samples, tutorials, reference architectures and other resources to help you to build your own ML@Edge (Machine Learning at the Edge) solution at AWS.

Tutorials/Reference implementations

Use Case	Description
How to detect anomalies in "Wind Turbines" in real-time?	This sample provides an end-to-end solution that manages the lifecycle of an anomaly detection model, deployed to a simulated fleet of wind turbine
How to classify images in the browser?	This sample provides an end-to-end solution that manages the lifecycle of an image detection model, deployed to a local device (laptop, mobile)

Contributing

If you have a question related to a business challenge that must be answered by an accelerated AI/ML solution, like the content in this repo, then you can contribute. You can just open an issue with your question or if you have the skills, implement a solution (tutorial, workshop, etc.) using Jupyter notebooks (for SageMaker Studio or Notebook Instances) and create a pull request. We appreciate your help.

Please refer to the CONTRIBUTING document for further details on contributing to this repository.

ml-edge-getting-started's Issues

samples/onnx_accelerator_sample1: missing instructions for installing dependencies

Describe the issue

I'm trying to follow the process explained in samples/onnx_accelerator_sample1/Greengrass.md, however during the step npx projen I receive the following error message:

❯ npx projen
👾 default | python .projenrc.py
Traceback (most recent call last):
File "/Users/<...>/development/ml-edge-getting-started/samples/onnx_accelerator_sample1/.projenrc.py", line 18, in
from projen.awscdk import AwsCdkPythonApp
ModuleNotFoundError: No module named 'projen'

after creating a new python venv and running pip install -r requirements.txt and pip install -r requirements-dev.txt it works.

It seems like the installation of the python dependencies is missing from the instructions?

Links

samples/onnx_accelerator_sample1/Greengrass.md

mqtt:subscription

Describe the issue

It should be $ mosquitto_sub -d -t turbine/raw instead of $ mosquitto-sub -d -t turbine/raw

Links

https://github.com/aws-samples/ml-edge-getting-started/tree/main/samples/onnx_accelerator_sample1/simulated_device

samples/onnx_accelerator_sample1: CodeBuild fails during build_deployment_package.py

Describe the bug

I'm running through samples/onnx_accelerator_sample1/Greengrass.md

After the model is approved, the CodeBuild project is triggered. However it fails with the following:

Traceback (most recent call last):
File "/codebuild/output/src3912502068/src/build_deployment_package.py", line 66, in
response = client_sm.describe_model_package(ModelPackageName=model_package_arn)
File "/root/.pyenv/versions/3.9.5/lib/python3.9/site-packages/botocore/client.py", line 508, in _api_call
return self._make_api_call(operation_name, kwargs)
File "/root/.pyenv/versions/3.9.5/lib/python3.9/site-packages/botocore/client.py", line 911, in _make_api_call
raise error_class(parsed_response, operation_name)
botocore.exceptions.ClientError: An error occurred (AccessDeniedException) when calling the DescribeModelPackage operation: User: arn:aws:sts:::assumed-role/onnxacceleratorsampleone-packageonnxmodelRoleFE70-4AT26K2V5LCR/AWSCodeBuild-15a8a3d5-ab63-4825-954f-127a14d295ef is not authorized to perform: sagemaker:DescribeModelPackage on resource: arn:aws:sagemaker:::model-package/modelPackageGroupTurbine/1 because no identity-based policy allows the sagemaker:DescribeModelPackage action

Expected Behavior

CodeBuild succeeds

Current Behavior

codebuild fails with IAM permission error

Reproduction Steps

run through the steps described in samples/onnx_accelerator_sample1/Greengrass.md

Once the model is approved the CodeBuild project will start and fail with the above error message.

Possible Solution

this seems to be caused by case sensitiviy missmatch between the actual model package name and the iam policy.

The model package name is modelPackageGroupTurbine but the IAM policy refers to the resource ...:model-package/modelpackagegroupturbine/*"

fix the CodeBuild IAM Role to match the

resourcarn:aws:sagemaker:::model-package/modelPackageGroupTurbine/1

Additional Information/Context

it seems that the lower case is intentionally created, there is a .lower() call when the IAM policy is created:

ml-edge-getting-started/samples/onnx_accelerator_sample1/onnxacceleratorsampleone/main_stack.py

Line 388 in fec319f

'arn:aws:sagemaker:'+ Aws.REGION+':'+ Aws.ACCOUNT_ID+':model-package/'+cfn_model_package_group.model_package_group_name.lower()+'/*' #the model group name needs to be lowercase

Sample version

No response

Region experiencing the issue

eu-west-1

Code modification

no modifications

Other information

No response

Service quota

• I have reviewed the service quotas for this sample

FEATURE REQUEST: Provide ready to deploy components to be used with AWS IoT Greengrass V2

Ready to deploy examples integrated with Greengrass v2