none
Example/sample of how to retrain Azure ML model using multiple input data sources RRS feed

Answers

  • If you just want to read multiple blob files and treat them as a single combined training dataset, using wild card as the blob URL for the Import Data module should just work. For retraining, you need to configure an Import Data module with that wild card URL. Then you simply add new data to the blob following that naming convention, and follow the retraining API doc. There is no other magic to it.
    Thursday, June 16, 2016 2:23 AM
    Moderator
  • Indeed there is sample code for what you are trying to do.

    1. Create your training experiment with multiple web service input modules

    2. Deploy the experiment as a web service

    3. On the resulting Web Service Dashboard, click on BES Help page link

    4. Go down the page to the C# sample code.

    thanks,

    Raymond

    Wednesday, June 22, 2016 4:20 AM

All replies

  • This should help.

    https://azure.microsoft.com/en-us/documentation/articles/machine-learning-create-models-and-endpoints-with-powershell/

    Wednesday, June 15, 2016 1:49 PM
    Moderator
  • Document on retraining process: https://azure.microsoft.com/en-us/documentation/articles/machine-learning-retrain-models-programmatically/

    Scheduled Retraining: https://azure.microsoft.com/en-us/blog/retraining-and-updating-azure-machine-learning-models-with-azure-data-factory/

    -Raymond

    Wednesday, June 15, 2016 3:37 PM
  • Hi Raymond,

    Thanks for replying, but I didn't find a code sample which showed multiple blob inputs during retraining. Can you please just point me to the code snippet or just post the snippet?

    I have tried all sorts of combinations

    - tried to use wildcard to read all the blobs from a container folder - doesn't work.

    - tried to add multiple inputs while building the JSON for the batch execution - fails.

    I wish Microsoft had published the API details. This shouldn't be hard.

    Best regards,

    Rahul


    Rahul.Kumar

    Wednesday, June 15, 2016 8:12 PM
  • Hello,

    As far as I can see in your retraining code, it uses single input. I am sure I am missing something. Can you point me to the code file name and line number please?

    Much appreciated!

    Regards

    Rahul


    Rahul.Kumar

    Wednesday, June 15, 2016 8:31 PM
  • If you just want to read multiple blob files and treat them as a single combined training dataset, using wild card as the blob URL for the Import Data module should just work. For retraining, you need to configure an Import Data module with that wild card URL. Then you simply add new data to the blob following that naming convention, and follow the retraining API doc. There is no other magic to it.
    Thursday, June 16, 2016 2:23 AM
    Moderator
  • Hi Hai,

    I am expecting that there is no magic to it. But there is massive lack of documentation. It is such a shame actually.

    I will appreciate if you can provide some precise reference to some code snippet, which shows a working solution of how to programmatically train a model using multiple blobs as input using wild cards.

    Wild card doesn't seem to work.

    The API seems to expect a filename with csv,/ tsv,/ arf extension.

    I have tried wild cards in both batch execution call as well as using ADF. It fails.

    Thanks

    Rahul


    Rahul.Kumar


    Thursday, June 16, 2016 8:16 AM
  • Hi, As Hai pointed out, 'Import Data' is the way to go. Are you using the 'Import Data' module in your experiment with Blob Storage as the data source?

    https://azure.microsoft.com/en-us/documentation/articles/machine-learning-data-science-import-data/

    Thursday, June 16, 2016 6:33 PM
  • Krishna,

    I am not trying to do this via Azure ML Studio where there is an 'Import Data' module.

    I am trying to do this programmatically. Is there an 'Import Data' module for this?

    Thanks for your suggestion, but this doesn't help.

    Rahul


    Rahul.Kumar

    Friday, June 17, 2016 8:00 AM
  • Indeed there is sample code for what you are trying to do.

    1. Create your training experiment with multiple web service input modules

    2. Deploy the experiment as a web service

    3. On the resulting Web Service Dashboard, click on BES Help page link

    4. Go down the page to the C# sample code.

    thanks,

    Raymond

    Wednesday, June 22, 2016 4:20 AM