none
Deploying non linear R models RRS feed

All replies

  • Pratik,
    Thanks for your question.  Indeed it is possible to deploy non-linear models.  There are several options available:
    1. With the MicrosoftML package you get rxFastForest and rxFastTrees.  The latter is also a fast-performance Boosted Decision tree package, similar to XGBoost. rxFastForest is a random forest implementation.
    Alternatively you can try rxBTrees, a boosted decision trees from RevoScaledR package.

    Once those models are trained, you can deploy them in either web service, SQL, or Spark

    Regards,

    Gleb


    MSFT

    Wednesday, May 9, 2018 3:11 PM
  • Hi Pratik,

    If your model is big it is recommended to use the dedicated pool feature in 9.3:

    https://docs.microsoft.com/en-us/machine-learning-server/operationalize/python/how-to-create-manage-session-pools

    https://docs.microsoft.com/en-us/machine-learning-server/operationalize/how-to-create-manage-session-pools 

    Friday, May 11, 2018 12:17 AM