Best practice when dealing with model reset

Ross Perry 1 Reputation point
2022-10-10T15:13:05.217+00:00

I need to increase my reward wait time to 24hours but doing so results in a reset of my model and a retrain using past 2 days of data.

I have been training my model for a long time now so I don't want it to forget months of data.

  • Does resetting my model bring it back to the beginning of the learning stage hence forgetting months of pre-learnt data?
  • Can I essentially 'save' my models hyperparameters by running an evaluation then applying the best policy after the reset?
  • Why is the past 2 days of data important?
Azure AI Personalizer
Azure AI Personalizer
An Azure artificial intelligence service that enables applications to personalize user experiences by learning from collective real-time user behavior.
32 questions
Azure AI services
Azure AI services
A group of Azure services, SDKs, and APIs designed to make apps more intelligent, engaging, and discoverable.
2,367 questions
{count} votes

1 answer

Sort by: Most helpful
  1. romungi-MSFT 41,961 Reputation points Microsoft Employee
    2022-10-12T09:07:42.633+00:00

    @Ross Perry I have some answers to your questions below from dev. I hope this helps!!

    • Does resetting my model bring it back to the beginning of the learning stage hence forgetting months of pre-learnt data?

    Resetting the model wipes out all trained model weights and policies, then retrains the model based on the past 2 days of data.

    • Can I essentially 'save' my models hyperparameters by running an evaluation then applying the best policy after the reset?

    Yes, or you can export the model policy and import it again after resetting the model

    • Why is the past 2 days of data important?

    2 days allows for training on the most recent data. Customizable retrain durations are part of the roadmap to allow more than 2 days data to retrain the model

    If an answer is helpful, please click on 130616-image.png or upvote 130671-image.png which might help other community members reading this thread.

    0 comments No comments