How to Find Azure ML Experiment based on Deployed Web Service

Question

I have a list of ML experiments which I have created in Azure Machine Learning Studio. I have deployed them as web services (the new version, not classic).

How can I go into Azure Machine Learning Web Services, click on a web service (which was deployed from an experiment), then navigate back to the experiment / predictive model which feeds it?

The only link I can find between the two is by updating the web service from the predictive experiment, which then confirms what the web service is. I can see that the "ExperimentId" is a GUID in the URL when in the experiment and the web service, so hopefully this is possible.

My reasoning is that relying on matching naming conventions, etc., to select the appropriate model to update is subject to human error.

Answer 1

The new web service does not store any information about the experiment or workspace that was deployed (not all new web services are deployed from an experiment).

Here are the options available to track the relationship between the experiment and a new web service.

last deployment

However, the experiment keeps track of the last new web service that was deployed from the experiment. each deployment to a new web service overwrites this value.

The value is stored in the experiment graph. One way to get the graph is to use the powershell module amlps

Export-AmlExperimentGraph -ExperimentId <Experiment Id> -OutputFile e.json

e.json

{
"ExperimentId":"<Experiment Id>",
// . . .
"WebService":{
// . . .
"ArmWebServiceId":"<Arm Id>"
},
// . . . 
}

azure resource tags

The tags feature for Azure resources is supported by the new web services. Setting a tag on the web service programmatically, with powershell or via the azure portal UX can be used to store a reference to the experiment on the new web service.