Skip to content

Latest commit

 

History

History
256 lines (195 loc) · 12.9 KB

README.md

File metadata and controls

256 lines (195 loc) · 12.9 KB
Raw

AIML21: Developers Guide to AI: A Data Story 

Train-the-Trainer Guide

Session Abstract

In this theatre session we will show the data science process and how to apply it. From exploration of datasets to deployment of services - all applied to an interesting data story. This will also take you on a very brief tour of the Azure AI Platform.

How to Use

Welcome, Presenter!

We're glad you are here and look forward to your delivery of this amazing content. As an experienced presenter, we know you know HOW to present so this guide will focus on WHAT you need to present. It will provide you a full run-through of the presentation created by the presentation design team.

Along with the video of the presentation, this document will link to all the assets you need to successfully present including PowerPoint slides and demo instructions & code.

  • Read the document in its entirety.
  • Watch the video presentation
  • Ask questions of the Lead Presenter

Assets in Train-the-Trainer Kit

Demo Instructions:

Please download all the code and data files in this repository to get started. There is a code and data folder which will contain all information you need to run all demos

You will need:

Setup Instructions:

  • Use the Deploy to Azure button below to create your Azure Machine Learning workspace to use throughout the session

Deploy to Azure

  • Log into the Azure Portal and your Azure Machine Learning Service
  • Launch the Preview UI
  • Create a Notebook VM
    • Select 'Notebook VMs' from left pane
    • Select 'New'
    • Provide a 'Notebook VM Name' (all lowercase)
    • Keep the default size VM provided
    • Once created select 'Jupyter Link'
    • Enter the user folder by double clicking
    • Select the upload button and upload the files listed below:
    • Open config.json and enter your subscription key, resource group and Azure ML Workspace name and save the file

Learn more about Notebook VMs here

  • Create a Compute Instance
    • Select 'Compute' from left pane
    • Select 'Add'
    • Provide a 'Compute Name' (all lowercase)
    • Select 'Machine Learning Compute' as Compute type
    • Choose a VM size
      • For standard compute select something like 'Standard_DS2_v2'
      • For GPU compute select 'Standard_NC6'
    • Select 'Low Priority' for Virtual machine priority
    • Minimum number of nodes set to 0 (then it will scale down completely and reduce costs)
    • Click 'Create'

Learn more about Azure Machine Learning Compute here

  • Upload the Dataset
    • Select 'Datasets' from left pane
    • Select 'Create dataset' and then 'from local files'
    • Select the 'Browse' button and find the data_train_experiment.csv file
    • Wait for the file to upload
    • Select 'Done'

Learn more about creating datasets here

  • Create a Power BI Workspace
    • Open powerbi.microsoft.com
    • Log in with your workplace credentials
    • Click on Workspaces on the left pane
    • Click Create a workspace
    • Enter a workspace name
    • Enter a description for the workspace
    • Click Save
  • Publish the Power BI Report
    • Open Power BI Desktop with Present.pbix
    • Click Publish
    • Select the workspace name you just created
    • Wait for the report to publish
  • View the Power BI Report
    • Go back to powerbi.microsoft.com
    • Refresh the browser if needed
    • See the report on the left pane in the web browser
    • Click on the report and it will load

Demo 1 - Explore

Video Demo with Voice Over: AIML21 - Demo 1 - Explore

  • Sign into your Azure Portal and open the Azure Machine Learning Workspace you created with the template above
  • Launch the Preview experience by clicking 'Launch Preview Now'
  • Open 'Notebook VMs' from the left pane
  • Select the 'Jupyter' Link
  • Double click the user folder
  • Log in if needed with your Azure Credentials
  • Open Explore.ipynb (recommend having all these tabs open and ready already)
  • Make sure all output cells in the notebook are cleared - Select Cell -> All Output -> Clear
  • Review the video for all talking points for each demo
  • Run through and explain the first 6 cells briefly
  • Note further cells and visualizations are available and attendees should download the code from aka.ms/AIML21repo

Demo 2 - Experiment

Video Demo with Voice Over: AIML21 - Demo 2 - Experiment

  • Go to ml.azure.com Azure Machine Learning Preview interface. You can access this via the URL directly and then sign in and select the correct workspace you created for this talk. Or from the Azure Machine Learning Service workspace in the portal selecting the 'Launch Preview now' button
  • Select Automated ML on the left pane
  • Review the video for all talking points for each demo
  • Click 'Create experiment'
  • Provide an experiment name, example/ aiml21-regression-duration
  • Select your Machine Learning compute you setup above
  • Select Next
  • Select the dataset you uploaded above
  • Review the preview of the dataset
  • Deselect the columns below:
    • SupportTicketID
    • CustomerID
    • DateCreated
    • DateCompleted
    • CustomerID2
    • CompletedTutorial
    • Rating
    • GeographyID2
    • RoleID2
    • DateCompletedDay
    • DateCompletedMonth
    • DateCompletedYear
  • Select Regression for Prediction Task
  • Select Duration for Target column
  • Select Advanced Settings drop down
  • Set Primary metric to 'normalized_root_mean_squared_error'
  • Set Training Job time (minutes) to 10
  • Select Start button

Once it is running, explain it will take some time to run the experiment and you will review a previously run example.

Please make sure you have a fully run Automated ML regression as explained about before getting on stage

  • Select previously run experiment from the Automated ML page from the left, or press the back button from the experiment you just submitted above
  • The previously run experiment will have a long 'Run Id' select this link
  • Review the video for all talking points for each demo

Demo 3 - Deploy

Video Demo with Voice Over: AIML21 - Demo 3 - Deploy

You will use the previously run experiment you reviewed above to deploy

  • Start on the screen with the run Automated ML experiment
  • Select the 'Deploy Best Model' button in the top right
  • Create a deployment name, example/ aiml21-v1
  • Add a deployment description if you wish
  • Choose Deploy button
  • Comment on the time of the deployment message in the top right corner

Please make sure you have a fully run Automated ML regression experiment that has already been deployed to an ACI before getting on stage

  • Select Endpoints on the left pane
  • Select your already created Azure Container Instance model
  • Review the video for all talking points for each demo
  • Select the browser tab with Jupyter
  • Open Deploy.ipynb
  • Make sure all output cells in the notebook are cleared - Select Cell -> All Output -> Clear
  • Confirm the config.json file is completed with the correct information
  • Edit the 2nd cell to include the name of your deployed web service
  • Run all cells in notebook and explain the code - Review the video for all talking points for each demo

Demo 4 - Present

Video Demo with Voice Over: AIML21 - Demo 4 - Present

Before getting on stage, open Power BI Desktop with Present.pbix and https://powerbi.microsoft.com/ with the workspace and report open from the setup above

  • Open Present.pbix
  • Review the video for all talking points for each demo
  • Show the Publish button
  • Open https://powerbi.microsoft.com/
  • Review the video for all talking points for each demo

Teardown Instructions

Full Teardown

  • Enter the Azure Portal and delete the Azure resource group to remove all resources for this project

To Save Costs

  • Make sure you select shutdown on your Notebook VM whenever not in use
  • When not in use delete the Azure Container Instance in endpoints

Resources and Continued Learning

Microsoft Learn:

Feedback Loop

All speakers will have the opportunity to attend Train-the-Trainer Q&A sessions every month (two sessions covering all timezones) and also after delivering the content on stage we would be keen for you to join us in the Feedback sessions organized to share your experience of the content and anything you would like to provide feedback or ideas on.

Do you have a comment, feedback, suggestion? Currently, the best feedback loop for content changes/suggestions/feedback is to create a new issue on this GitHub repository. To get all the details about how to create an issue please refer to the Contributing docs

Become a Presenter

To become a certified presenter, contact scalablecontent@microsoft.com. In your email please include:

  • Complete name
  • The code of this presentation: aiml21
  • Link to a video of you presenting (~10 minutes in length)(ex: unlisted YouTube video).

    It doesn't need to be this content, the importance is to show your presenter skills

A mentor will get back to you with information on the process.

Certified Presenters

Thanks goes to these wonderful people (emoji key):

Amy Boyd
Amy Boyd
📢 📖