Retrain Classic Machine Learning Model in Data Factory RRS feed

  • Question

  • Dear Microsoft,

    I am really struggling to figure out how to retrain a machine learning model from ML Studio in Data Factory, and have not found much information relevant to doing so in the Version 2 UI. I cannot join the dots between tutorials I have seen using V1, and the V2 user interface with drop in modules for Batch Execution and Update Resource Activity.

    Could you advise me on where I can get some help with understanding how to do this? It cannot be a difficult task, otherwise people wouldn't do it, I just can't seem to understand what exactly I should be doing.

    Also, I am currently using a free trial so I do not know if some of the Data Factory functions are unavailable, please let me know if this is the case - I will soon be moving to a full subscription. 

    Thank you for any support you can provide!

    Wednesday, September 11, 2019 2:15 PM

All replies

  • Hello New_To_Azure and thank you for your inquiry.

    The Data Factory activities work on published web services.  You will need 3 of them:

    1. Scoring
    2. Retrain
    3. Update model

    Retrain model and update contains two steps in AML studio: retrain the model with a web service, which would created an ilearner file (using batch execution), and then use the update resource activity to update model with the created ilearner file. So retrain and update should have different web service in AML Studio

     You can retrain a model from a Data Factory pipeline by doing the following steps:

    1. Publish the training experiment (not predictive experiment) as a web service. You do this step in the Azure Machine Learning studio as you did to expose predictive experiment as a web service in the previous scenario.
    2. Use the Azure Machine Learning studio Batch Execution Activity to invoke the web service for the training experiment. Basically, you can use the Azure Machine Learning studio Batch Execution activity to invoke both training web service and scoring web service.

    After you are done with retraining, update the scoring web service (predictive experiment exposed as a web service) with the newly trained model by using the Azure Machine Learning studio Update Resource Activity. See Updating models using Update Resource Activity article for details.

    In the 'Azure ML' tab of the ML Batch Execution and ML Update Resource activities, the AML Linked Service is the published web service you want to operate on.

    Hope this info helps. If you have any further query, please do let us know.

    Tuesday, September 17, 2019 1:16 AM
  • Hi there,

    Just checking to see my previous comment was helpful? If you have any further query, do let us know.

    Thursday, September 19, 2019 7:13 PM
  • Since I have not heard back from you, I will assume you found your own solution.
    Monday, September 23, 2019 5:32 PM