Idea is to demonstrate data science workflow where:
- User can work with local developer laptop to explore and model
- User can use any open source ML toolkit, in this example it is scikit-learn
- Model can be trained efficiently with remote (GPU) servers with same code
- Trained model can be deployed in continous fashion into as a rest endpoint
- Create a new Machine Learning Service Workspace in portal.azure.com (or use a shared one)
- Download the configuration ('download config.json' from overview pane)
- This was developed with Python 3.7
- Download the MNIST dataset by running
python initial_setup
.