Custom trained model in Azure ML

Question

I have a predictive model which I trained on a training set. I have written it in R. Now I want to deploy it as a web service so anyone can just input the data into it and get the output from the predictive model.

I wanted to use Azure ML for deploying. I wanted to know whether I can drag and drop my already created/custom trained model to Azure ML studio instead of re-training it there ? I know we can train and save models in AML Studio but I am not sure about adding already created models and using them in AML solution. Help regarding this will be appreciated.

Answer 1

No you cannot. I had a discussion with someone in their development/support team on the MSDN forums and currently they don't support 'drag and drop' type of functionality. However you CAN serialize the model output and then de-serialize them in Azure.

Note that the answer in the image is a bit outdated and there is the 'Create R Script' module to replace the serialization-deserialization steps within Azure. However I believe you can still serialize outside Asure (in your Desktop version of R) and deserialize them in Azure.

Link to the conversation in Image: https://social.msdn.microsoft.com/Forums/azure/en-US/5944c342-79ac-4ada-8006-8edf40f36ee1/r-script-as-a-trained-model?forum=MachineLearning

Answer 2

I don't think you can currently upload a trained model.

Your options would be to either re-train the model in AzureML or expose them as a web-service using an Azure Virtual Machine running something like:

• Rook
• Shiny
• DeployR

Answer 3

Look at using Power Query. For example: http://www.poweredsolutions.co/2014/11/06/getting-api-results-from-azure-machine-learning-into-power-query/ As per the bottom of the url, power query can be a source into AML.

Answer 4

It appears that this can now be done to some degree with the AzureML Package.

See the next to last section, labeled "Other examples of publishing web services" on this page: https://htmlpreview.github.io/?https://github.com/RevolutionAnalytics/AzureML/blob/master/vignettes/getting_started.html