none
Getting multiple serialized trained models in execute R script as its output RRS feed

  • Question

  • I need to do a conformal prediction which needs to train two models. The first can be any type of ML model and the second one is training an error model using the best ML model trained. I have been training these two models in R locally and then upload them in AZURE ML to create a prediction experiment with its corresponding web API. Unfortunately, the size of the files are huge and the loading process is very slow and anytime I want to do the prediction, the loading process is refreshed. I wanted to do that directly in AZURE ML. I could train these two models in Execute R Script but unfortunatly I can only have one of them as an output of the module and use it in a different Execute R script for prediction purposes. I know I can use the new Create R model module to train my models there but the same problem exists for the scoring. I appreciate if someone can help me understand how I can extract more than one data/model from Execute R Script.

    Thanks

    Tuesday, December 4, 2018 6:27 PM

All replies

  • Hi, 

    Perhaps other users in the forum might be able to help in here as well. I am wondering if a 'Custom R Module' might help? https://docs.microsoft.com/en-us/azure/machine-learning/studio/custom-r-modules

    Regards,
    Jaya

    Tuesday, December 4, 2018 9:30 PM
  • Hi. Thank you for your response. Unfortunately the R package I need is not compatible with the built in caret package in R in the Azure platform and the "custom R model" module dos not let us upload our own packages there so it is not an option. However I found a post using a serilized list of models from R script in Azure which can be an option but since boosted decision trees models are huge, it did not work out for my case.  I ended up training my model in R and uploading the trained models as Zip files in Azure. Although it works fine, it is very slow.
    Thursday, December 13, 2018 2:54 PM