IssueTracker

Please describe your requested enhancement. Good feature requests will solve common problems or enable new use cases.

What you would like to accomplish:
Reduce the resources and costs regarding AutoML models deployed in standby. A new approach to current service features.

How this might work:
A self-managed autodeployment on Google side that takes care of the deployment and undeployment.
Whenever I make a REST call to my AutoML Vision model, the model is deployed (processes the prediction and provides the response) and after a while (without any other prediction requests, inactivity) it is automatically undeployed again until the new prediction arrives. Of course the first prediction would have some extra minutes delay in the response but the following ones during the selected period would be processed normally. When the model becomes idle for a period bigger than the selected one, it automatically undeploys itself.


If applicable, reasons why alternative solutions are not sufficient:
Right now it needs to be monitored and do it manually.