Can we train Azure virtual machine Scale sets predictive auto-scale model with our past data.

Question

The new feature of azure VMSS is predictive auto-scaling using a inbuilt ML- model and trainig with the real time data , it starts prediction after a minimum of 7 days of deployment.

So my question is this , can we provide our previous data, suppose our past 1 year data to train this model according to one's need and after training and also with daily usage of our cloud usage patterns, will we be able to auto-scale?

Answer 1

@Kushal JAIN

I understand you wish to use your past data to train the predictive auto-scale model. Currently predictive autoscale does not support the experience where users can feed historical data. If you want to view the accuracy of the data, you can turn on Forecast only. Forecast only allows you to view your predicted CPU forecast without triggering the scaling action based on the prediction. You can then compare the forecast with your actual workload patterns to build confidence in the prediction models before you enable the predictive autoscale feature.

Hope this answers your question. I have passed on your feedback to the autoscale product team regarding this feature.

---

Please don’t forget to "Accept the answer" and “up-vote” wherever the information provided helps you, this can be beneficial to other community members.