| | 1

Error while creating a web service for a predictive experiment in Azure Machine Learning Studio (AzureML)

For a project for the detection of the content emails, I’ve created anAzureML experiment. But it was not possible to create a web service from the predictive experiment. After several minutes, Machine Learning Studio reported the following error:

Service call failed due to a network error. 
Error when requesting https://europewest.studioapi.azureml.net/api/workspaces/017690d3ae8c44829bbxxxxxxxxxxxxxxxxxx/webservicegroups/xxxxxxxxxxxxxxxxxxxxxxxxx/modelpackages/ 
Request opened: 2016-04-13T14:48:16.449Z Onerror called: 2016-04-13T14:48:21.843Z

It took me days to find out why. In my case, it’s because that the function “Feature Hashing” creates a large number of additional columns in the data stream (matrix). With the high number of columns, the web service cannot be created. The problem can be circumvented if you filter the columns with “Project Columns”, so that only the necessary fields are propagated to the web service output (and not just the feature matrix):