Training model using Azure Machine Learning and Jupyter Notebook VM

What you will do

• Create a Notebook VM to your Azure Machine Learning Workspace
• Train a model using Azure Machine Learning and Notebook VM. Deploy the model to your Vision AI Dev Kit device

Pre-requisites

• Vision AI DevKit camera
• Azure Machine Learning workspace

Create Notebook VM

• Log in to Azure Portal and access Azure Machine Learning workspace
• Create a new Notebook VM
• 

Clone example notebooks from GitHub

• Add an access token to your GitHub account (if not exist yet). This is needed for being able to clone a private GitHub. In this case the password when cloning a GitHub is your token.
• Open the Notebook VM
• Open “Terminal”
• Note that if you use Edge as your browser the terminal many not open directly. In that case you can access it from “Running” tab.

git clone --recursive https://github.com/microsoft/vision-ai-developer-kit

Prep your environment

• Change the aml_config/config.json file to match the subscription details for your Azure ML Workspace
• In case you want to use your own data go to the data folder and create a folder with your data, e.g. my_data.
  • In the my_data folder, copy your data. Each folder name is the label of the images in that folder. For example, the soda_cans folder looks like this:
    • soda_cans
    • coke
    • ice
    • pepsi
• IMPORTANT: You MUST upload more than 35 images per folder so there’s enough data for training.

Run the notebook

• Open the 02-mobilenet-transfer-learning-final.ipynb notebook

• Set the Kernel to Python 3.6

• Run the notebook by running each cell individually

• In case you don’t want to run a cell you can change it to markdown

Configure your IoT Hub details

• Change the cell content to reflect your IoT Hub. Please note that you can use format:
  • Iot_hub_name=”youriothub”