ResponsibleAI-Airlift-July2020

Overview

This repository contains content of a two-part workshop for using machine learning interpretability and fairness assessment (+ unfairness mitigation) to build fairer and more transparent models. The different components of the workshop are as follows:

• Part 1: Interpretability with glassbox models (EBM)
• Part 2: Explain blackbox models with SHAP (and upload explanations to Azure Machine Learning)
• Part 3: Run Interpretability on Azure Machine Learning
• Part 4: Model fairness assessment and unfairness mitigation

Getting started with the workshop environment

1. Provision your personal Lab environment

   • Open Registration URL: http://bit.ly/2OjknZW
   • Enter Activation Code which should be provided by the instructors of the workshop.
   • Fill out registration form and Submit it.
   • On the next screen click Launch Lab.
   • Wait until your personal environment is provisioned. It should take approximatly 3-5 minutes.

2. Login to Azure ML studio

   • Once the workshop enviroment is ready, you can open new browser tab and navigate to Azure ML studio, using it's direct URL: https://ml.azure.com. We recommend to use Private Browser window for the login to avoid conflicting credentials if you already have Azure subscription.
   • Use credentials provided in the workshop environment to sign-in to Azure ML studio.
   • In the Welcome screen select preprovisioned subcription and workspace similar to screenshot below:
   • Click Get started!
   • In the welcome screen click on Take a quick tour button to familiarize yourself with Azure ML studio.

3. Create Azure Machine Learning Notebook VM

   • Click on Compute tab on the left navigation bar.
   • In the Notebook VM section, click New.
   • Enter Notebook VM name of your choice and click Create. Creation should take approximately 5 minutes.

4. Clone this repository to Notebook VM in your Azure ML workspace

   • Once Notebook VM is created and in Running state, click on the Jupyter link. This will open Jupyter web UI in new browser tab.
   • In Jupyter UI click New > Terminal.
   • In terminal window, type and execute command: ls
   • Notice the name of your user folder and use that name to execute next command: cd <userfolder>
   • Clone the repository of this workshop by executing following command: git clone https://github.com/microsoft/responsibleai-airlift.git

5. Open Part 1 of the workshop

   • Go back to the Jupyter window.
   • Open Interpretability with glassbox models (EBM) notebook.

You are ready to start your workshop! Have fun.

Useful Links

Interpretability

• InterpretML open source
• InterpretML AI Show
• Interpretability in AzureML, concept
• Interpretability in AzureML, how-to
• InterpretML case study

Fairness

• Fairlearn open source
• Fairlearn whitepaper
• Fairlearn AI Show
• Fairness Assessment in AzureML, concept
• Fairness Assessment in AzureML, how-to
• Fairlearn case study