Provide Pip wheels for common Python packages
Installing Pip packages that require system binaries, like Scipy and Numpy, is a headache on Azure because you currently have to set up a local Win32 Python environment and compile your wheels manually. Azure should provide a PyPI server that has pre-built wheels for all the common Python packages for the architecture on Azure web apps.
We have added a new feature called the Notebook Service that allows launching Jupyter with Python 2 and 3 support from Azure ML.
We also released a new feature that enables publishing models and functions as web services directly from Python - running locally (e.g. on Jupyter). The following blog post provides an example on how to use this feature.