none
Automated Machine Learning Container Deployment Service Suddenly Returning "'y_query" Error RRS feed

  • Question

  • I was recently able to use the Azure Automated Machine Learning service to deploy a container instance for my model and receive predictions via C#/REST.  However, I recently retrained my model and redeployed and now the service is returning:

    ""{\"error\": \"'y_query'\"}"

    Any idea why this might be happening all the sudden? 

    How can I get more information to resolve this error?

    Thursday, October 3, 2019 4:22 PM

All replies

  • Hello,

    Could you please check if the compute that is attached to this deployment is up? You can check the same through the deployments tab from azure portal and check the compute from compute tab if there are any valid nodes.

    Friday, October 4, 2019 10:26 AM
    Moderator
  • Hi Rohit,

    Sorry, I've deleted the compute thinking that it would not be necessary after initial training. 

    Are you saying the compute must be running in parallel with the deployment container?  Isn't the deployed container image "containerized" such that it should be standalone once it has its model?

    If not, is there a way for me to keep my costs low when running both a compute and an ACI container?  I find that running these computes can be very expensive.

    Thanks,

    Dom

    Friday, October 4, 2019 5:55 PM
  • Hello Dom,

    While deploying the web service using any compute type the minimum and maximum number of nodes are defined which will help to create a container when required. If the configuration is deleted after setting up the service the request would fail. You can take a look at this page on setting up ACI containers for web service. 

    -Rohit

    Thursday, October 10, 2019 7:04 AM
    Moderator