Changes to the Web Service publishing process (Operationalization)

Question

With the new release of Azure ML features, we are updating the web service publishing workflow. Below are the list of changes:
Web Service Input / Output modules: these replace the input and output ports previously used to create web services. You can now drag and drop them on the experiment from the left pane under Web Services (instead of right-clicking on the module).
Generation of Scoring experiment from Training experiment: We have automated the steps in creating a Scoring experiment by adding the Web service input and output modules, and automated the saving of the Trained model. With this feature, you no longer need
to manually save a trained model and add it to a Scoring Experiment. The "Create Scoring Experiment" button will trigger these events and generate the Scoring Experiment for you. The resulting Scoring experiment can be published as a web service
as is, or user can make modifications to the generated workflow including location of input/output modules.
To use this feature:
  ○ Create a Training Experiment (includes an algorithm such as Boosted Decision Tree and a Train Module)
  ○ Click Run
  ○ Click Create Scoring Experiment
  ○ Review the resulting Experiment, then Click Run
  ○ Then click Publish Web Service
A sample Training Experiment:
Publishing a Training experiment as a web service:Similar to the Scoring experiment, we have now automated the steps to create a web service from the Training experiment using the "Prepare web service" command button. This action adds Web service
input and output modules to the training experiment. The experiment can then be published as a web service. This  approach to creating a web service is needed for setting up the Trained model for retraining using the Retraining APIs. But is not the recommended
approach for creating web services - Scoring experiment should be used for that purpose.
Publishing an experiment as a web service: in cases where the experiment does not involve training, such as when processing, transforming and cleaning up data, it can also be published as a web service using the "Prepare web service" button.
We will publish more detailed documentation for this and other features shortly. Please let us know if you have any questions.
Thanks,
The Azure ML Team

Answer 1

Hey Freppas, can you give some more information about your first ask? For classifiers, we already output the scored probability along with the scored label. You can limit the output to just the probability using the Project Columns module to exclude unwanted
columns. The web service is not concerned with domain-specific functionality - it will output exactly what you see being output in Studio when you set up the web service ports
The second ask is something we're working on bringing to our product - feature importance. Again, you would have to make it happen in the Studio before you publish the web service for the web service to have that functionality