Introduction to a Basic Machine Learning Workflow for Predicting Materials Properties

2022.09.13 Benjamin Afflerbach, Materials Science and Engineering, University of Wisconsin-Madison

To run the Machine Learning Lab tool used in this presentation see: https://nanohub.org/tools/intromllab

This video is part of the Back to School Webinar Series on Teaching found at https://nanohub.org/groups/chem/live_...
Table of Contents available below.

This tutorial will introduce core concepts of machine learning through the lens of a basic workflow to predict material bandgaps from material compositions. As we progress through this workflow we will highlight key steps, challenges that can come up with materials data, and potential solutions to these challenges. The core workflow we’ll introduce includes: Data Cleaning, Feature Generation, Feature Engineering, Establishing Model Assessment, Training a Default Model, Hyperparameter Optimization, and Making Predictions. By the end of the tutorial I hope that you’ll have a better understanding of these core concepts, and how they can all fit together.

Table of Contents:
00:00 A Basic Workflow for Predicting Materials Properties
02:05 Summary
03:49 An Application: Predict a Materials Property
05:06 A Basic Materials Design Workflow
06:45 Machine Learning for Pattern Matching
08:03 Key Distinction in ML
09:14 Key Distinction in ML
10:45 Model Types
11:12 Decision Trees: Structure
12:07 Decision Trees: Inputs
14:22 Decision Trees: Outputs
15:22 Summary
17:02 Machine Learning Lab Module Demo
22:34 1. Data Cleaning and Inspection
30:22 2. Feature Generation
35:36 3. Feature Engineering
40:23 4. Setup for Model Evaluation
43:36 5. Fitting and Evaluating a Default Model
48:23 6. Improving the Model by Optimizing Hyperparameters
51:36 7. Making Predictions
53:16 Questions

This resource and related downloads can be found at: https://nanohub.org/resources/36471