Download a trained ML Model from Azure ML studio to deploy on a standalone computer

0
votes

I have setup a ML model in Azure ML studio and I am able to use the ML Studio's Web API to obtain predictions.

The key challenge with keeping the model hosted within Azure ML Studio is client computer's internet dependency and latency associated with each prediction.

I wanted to understand if there was any way to download a model created in Azure ML studio and consume it locally(say using a local .Net application).

This question was asked few years back, some people believed it was possible but without much details.

When a model has been trained, I do see the option of 'Save as Trained Model' as shown in the snapshot but I am not sure how this could eventually be downloaded to a local computer and consumed locally enter image description here

Looking forward to hearing potential solutions.

1
azuremachine-learningazure-machine-learning-studioazureml

1 Answers

1
votes

I think you are using the ML Studio classic version.

The Classic ML Studio doesn't support CODE SDKs.

enter image description here

You could migrate the newer Azure ML.

  • Migrate datasets to Azure Machine Learning.

  • Use the designer to rebuild experiments

. - Use the designer to redeploy web services.

Once you migrate, the model

enter image description here

The model gets outputted to blob

enter image description here

You can download the model from here.Not sure whether this will help you directly .Alternatively, you can use the trained model of the workspace by following the article here. The newer version supports the SDK. You will have to train and register the model in the ML Designer. Post which you can consume them from the local machine.