Tutorial 4A: Scoring Pipeline Deployment Introduction

By the end of this tutorial, you will predict the cooling condition for a Hydraulic System Test Rig by deploying a MOJO Scoring Pipeline into production using Driverless AI.

Key:

Complete
Failed
Available
Locked
Objective
Select the "Read" button to begin.
Select the "Read" button to begin. Explore the goal of this tutorial
Prerequisites
Select the "Read" button to begin.
Select the "Read" button to begin. Find general information about what you need in order to complete this tutorial
Task 1: Tour of Prebuilt Experiment
Select the "Read" button to begin.
Select the "Read" button to begin. There are different Driverless AI experiments you could build with the Hydraulic System data and each label, which include cooler condition, valve condition, pump leakage, accumulator gas leakage, and stable flag [1]. For each experiment we could build, Driverless AI would choose a classification scorer because each of the labels have 2 or more categories.
Task 2: Scoring Pipeline Deployment Concepts
Select the "Read" button to begin.
Select the "Read" button to begin. In the scoring pipeline deployment concepts section, we cover the general machine learning workflow, model training, production deployment, edge inference, Driverless AI MOJO Scoring Pipelines and Python Scoring Pipelines, how to choose a scoring pipeline, productionizing scoring pipelines, challenges of productionizing scoring pipelines and making decisions based on a good model.
Task 3: Batch Scoring via Score Another Dataset
Select the "Read" button to begin.
Select the "Read" button to begin. You will learn to use Driverless AI’s Score Another Dataset button to perform batch scoring on data. Behind the scenes, Driverless AI uses the Python Scoring Pipeline to do batch scoring. You will download the batch scores csv file and import it into your favorite spreadsheet software, but in the tutorial, we import it to Google Sheets.
Task 4: Interactive Scoring (REST and AWS Lambda) via Deployment Templates
Select the "Read" button to begin.
Select the "Read" button to begin. You will learn to use Driverless AI’s Deploy (Local & Cloud) button to deploy the MOJO Scoring Pipeline to a REST Server and also an Amazon Lambda. Then once the MOJO is deployed into production, you as a client will use curl to send data and the prediction function you want the MOJO to execute to make real-time predictions on your data. When the MOJO finishes making the predictions, it will send the results back to you.
Task 5: Challenge
Select the "Read" button to begin.
Select the "Read" button to begin. You will be provided with some project ideas on how you can go further with your Scoring Pipeline Deployment knowledge. You can deploy a new scoring pipeline based on a new dataset, integrate a scoring pipeline directly into an existing program or deploy the scoring pipeline using a Driverless AI deployment template.
Next Steps
Select the "Read" button to begin.
Select the "Read" button to begin. After you complete the tutorial, we provide you with extra resources on what to do next. If you want to further your learning on scoring pipeline deployment you can go to our next tutorial, check out some H2O.ai webinars on this topic or check out what other companies or people are saying about using Driverless AI scoring pipelines.
Appendix A: Build Experiment
Select the "Read" button to begin.
Select the "Read" button to begin. You will import the Hydraulic System data using a Driverless AI Data Recipe, split the data into a training and test set, configure the Driverless AI experiment and then launch the experiment so that you can be ready to use either of the Driverless AI scoring pipelines to make predictions and more.
Appendix B: Glossary
Select the "Read" button to begin.
Select the "Read" button to begin. If there were terms in the tutorial that you want more clarification on like how it is relevant to H2O-3 or Driverless AI or a general definition, we provide you the link to the AI Glossary in H2O.ai Community.
Quiz
20 Questions  |  2 attempts  |  16/20 points to pass
20 Questions  |  2 attempts  |  16/20 points to pass
Badge
  |  Badge available
  |  Badge available After passing the quiz, you can get your badge by clicking on "Badge Earned". Please check your email address for instructions on how to view, manage, and share your new badge!
You must be logged in to post to the discussion
  • RS

    Question no. 14: missing or misleading:

    14. What is not a part of the 4 pillars of ML model deployment?

    Data Sources and Their Format
    How To Measure Success
    Bringing the Model To Production
    Internal Limitations and Restrictions

    ALL of them ! (https://training.h2o.ai/webina...):

    Challenges of Productionizing Scoring Pipelines

    At this stage, the organization has already built their model(s) that deliver high accuracy and does not overfit. However, they have not obtained any economic value from the scoring pipeline because it is not used in production making decisions. For many organizations deploying the scoring pipeline into production is challenging because there are many paths and questions that must be considered. So in effort to help the organization find their path to model deployment, we have "The Four Pillars of ML Model Deployment". These pillars include points to think about as we move toward deploying our scoring pipeline into production.

    Figure 8: The Four Pillars of ML Model Deployment


    Reply
  • SP

    Thank you for the feedback Rino. As I stated in my email, the question is not wrong because the answer to that question is "Internal Limitations and Restrictions." The fourth pillar is about "EXTERNAL Limitations and Restrictions. " NOT "Internal Limitations and Restrictions."

    Reply
  • RS

    Task 2: wording to be completed:

    "...train the Machine Learning Algorithm and save the Machine Learning Model until you obtain the most accurate model possible".

    Reply
  • SP

    Thank you for the feedback. It will be incorporated to the tutorial. 

    Reply