DevOps for Machine Learning (Azure MLOps Part 5) - Deploy Your Model to Azure Kubernetes Service

In this series of videos I'm showing how to get started with DevOps for Machine Learning (MLOps) on Microsoft Azure.

In the fifth video of this 5-part series, you'll discover how to deploy your Machine Learning Model as web-based API in an Azure Kubernetes Service, as well as run Integration Tests again.

And don't forget to click the bell so you don't miss anything.


📢 If you enjoyed this video help others enjoy it by adding captions in your native language -

🚀Recommended links to learn more about DevOps for Machine Learning (MLOps):

✅ Let's connect:

DISCLAIMER: This video and description contains affiliate links, which means that if you click on one of the product links, I’ll receive a small commission. This helps support my channel and allows me to continue making awesome videos like this. Thank you for the support!

#MLOps #DevOpsForMachineLearning #AzureML

[maheshjoshi]

Hi Sascha, I followed this 5 step tutorial and able to achive MLOPs end to end pipeline.Thank you for such informative and complete tutorial.

[luizclaudios]

Congratulations!! Excelent videos and material. Tanks

[temiwale]

Thank you so much Sascha! You've really blessed me with this series. God bless You!

[gtosXD]

Really the best tutorial about MlOps on Azure!

[anirbansaha]

The series has been really great. This is exactly what I was searching for.

[parinaz]

Thanks for your great videos and material. You are absolutely adding value and advancing the knowledge of using devops in society...Microsoft should be proud....I am looking forward to see your ML from scratch in MLOps if it is something in your pipeline...cheers...

[rahulpbabbu]

Great content, thanks a lot for creating this wonderful series. Extremely helpful. Cheers!

[marialuoana]

great videos Sascha, really helpful and well explained, please do more in the future :)

[homarghyahomroy]

Thanks, Sascha, for your prompt reply as always. I think I am pretty close to finishing this 5 video series. In the "Deploy to Production" pipeline while I am trying to create the Kubernetes service I am getting the below error
"Provisioning compute resources...
2020-09-24T06:52:36.4981948Z Resource creation submitted successfully.
2020-09-24T06:52:46.9406090Z Provisioning operation finished, operation "Failed"
2020-09-24T06:52:47.0576070Z {'Azure-cli-ml Version': '1.14.0', 'Error': ComputeTargetException:
2020-09-24T06:52:47.0577859Z Message: Compute object provisioning polling reached non-successful terminal state, current provisioning state: Failed
2020-09-24T06:52:47.0578620Z Provisioning operation error:
2020-09-24T06:52:47.0586932Z InnerException None
2020-09-24T06:52:47.0587329Z ErrorResponse
2020-09-24T06:52:47.0587649Z {
2020-09-24T06:52:47.0587986Z "error": {
2020-09-24T06:52:47.0599373Z }
2020-09-24T06:52:47.0599528Z }}
2020-09-24T06:52:47.1621559Z ##[error]Script failed with exit code: 1".
I have tried with different VM size it seems the error is not going away. I think I need your expert advice.

[mathieuhervin]

Hi sascha dittman, thanks very much for your video series. They are very insightfull and helped me a lot. I got an error at the step of "Deploy ML Service to AKS" with the following error message:
"error": {
2022-03-02T10:29:20.7184193Z "message": "Conflict of operation, another operation on same entity is already running in workspace MLOps-tuto-ws."
2022-03-02T10:29:20.7184637Z }

But i dont have obviously any other running operation in this workspace. How would you advice to solve this? Thanks for your help

[chandanbilvaraj]

Hi Nice tutorial, I got an idea of how prod deployments should be done.

In create AKS Step if the AKS exists already i'm getting the below error:
A compute with the same name already exists. Updating property: provisioningState for compute is not supported yet
In the tutorial you mentioned that it would step over but its throwing an error. How can it actually step over?

Also Integration test is returning below error:
ERROR: -o/--override-ini expects option=value style (got: 'tsv)')

Command:
pytest integration_test.py --doctest-modules --cov=integration_test --cov-report=xml --cov-report=html --scoreurl $(az ml service show -g cbt-mlops -w mlops-wps -n diabservc-aks --query scoringUri -o tsv) --scorekey $(az ml service get-keys -g cbt-mlops -w mlops-wps -n diabservc-aks --query primaryKey -o tsv)

[reubenhilliard]

Hey Sascha, would you consider making a "Part 6" video that relates to everything monitoring, triggers, model drift -> automated retraining, network/usage/demand spikes -> "elastic" prediction scaling. That would be amazing!

[assadullah]

Hi
when i click the score url after aks deployment i get this message "Unauthorized, no Authorization header". can you please help me to resolve this issue. Thanks

[zohrehdoosti]

Thank you so much, Sascha for these awesome series of videos.
Can you share an example that shows the retraining of ML model happens after some metric performance changes? For example, if the model results based on real-time feed data is not good, a pipeline will be automatically triggered, and a new ML model will get retrained and deployed.

[bouhamidikaoutar]

With a student account is it possible to create the aks cluster because I have an error message saying that the cluster should have at least 12 cores to operate. So please if the is a way to do it, tell me how. Thanks

[MrMidhunjose]

liked the videos
I have seen all the 5 of the series on suggestion is please be more detailed in the scripts.

for eg when we deploy the ML Model Please ensure that each and every word in the Azure Ml commandlet is explained. need to know how each is working

thanke

[martin.thogersen]

If some of the DevOps parameters are changed, the changes will not be applied to the resources, because the resources already exists, right?

[romanborovikov]

Hi Sascha, thank you for the great set of videos. I didn't see a step in the pipeline to check a score of the re-trained model (if it's better than the deployed one)... Did I miss something? If there is no such a step how would you to implement that?

[temiwale]

Is anyone else getting this error on the Deploy to ML Service to AKS step: "Message: Compute resource with Id: aks is not in Succeeded state. Compute provisioning state: Failed"

[bouhamidikaoutar]

Please I have an error when I want to create azure kubernetes service it says the process '/bin/bash' failed with exit code 1.Thank you for the videos.