locked
Saving / loading trained models RRS feed

 > 

  • Question

  • Question
    Sign in to vote
    0
    Sign in to vote

    Hi,

    I'm building a service where we'll be doing lots of training models using the same workflows. I've seen the various posts on programmatic retraining of models and creating new endpoints etc, but I do wonder why it's so complicated. Would it not be simpler just to be able to specify a name for a saved model, save it, and then load it up again on the fly? I'm sure there's a reason for the additional complexity, I just don't know what it is. Please can someone enlighten me? (Or maybe tell me I'm wrong, and you can save / load models programmatically!)

    Thanks, James.

    (and sorry if this has been asked and answered - the forum search seems to be broken today)

    Thursday, October 1, 2015 9:46 AM

Answers

  • Question
    Sign in to vote
    0
    Sign in to vote

    Hi James,

    Saving the model and loading it on the fly is what Azure ML does by default. No coding needed there.

    You can also keep tweaking your model and updating your web service in the UI without needing to deal with retraining APIs.

    The retraining process is needed when you have a number of customers. They need to train the model with their own data which results in a version of the model trained for each of them. That also means they cannot share one web service endpoint - they each needs a web service endpoint with their version of the model. That is where the APIs come in - and some of the complexity.

    Having said that, we are working on providing a simplified interface for setting up retraining. Feel free to share your specific scenario with us so we can see if any needless complexity impacting you can be also resolved.

    thanks,

    Raymond

    • Proposed as answer by neerajkh_MSFT Monday, October 5, 2015 11:14 PM
    • Marked as answer by neerajkh_MSFT Saturday, October 17, 2015 7:40 PM
    Sunday, October 4, 2015 12:25 AM