PyMKS Overview

• PyMKS is a materials informatics toolkit (Materials Knowledge System) written in Python

• PyMKS provides a machine learning technique unavailable in other machine learning libraries

• The core of PyMKS is a set of convolution kernels, the rest is an API that works with Scikit-Learn

• PyMKS provides documentation and testing for MKS use cases

</p>

Tutorial

• live coding - follow along - lots of typing

• should have software installed - running a Jupyter notebook

  • install on your laptop

  • use Matin

  • use Binder

• please stop me if I go too fast

• feel free to ask questions

Syllabus

1. Python Intro?
2. Composing microstructure into a digital signal
3. Classify a microstructure using 2-point stats
4. Learn a Cahn-Hilliard simulation using regression

Tools: PyMKS, Scikit-Learn, Dask, Numpy

1. Python Intro

• How to use the Jupyter Notebook
• How many know Python?
  • Variables
  • Containers
  • Loops
  • Conditionals
  • Arrays, vector calculations
  • Plotting using Matplotlib

2. Composing microstructure into a digital signal

• Microstructure function
• Why do we need to generate features for each data set?
• Generate features for each sample of data
• Using PyMKS microstructure functions
  • continuous data
  • discrete data
  • periodic data
  • multiple variable data

3. Classify microstructures using 2-point stats

• Generate artificial microstructures
• Discretize the data using a discrete microstructure function
• Create features for each sample using 2 points stats
• Dimensionality reduction on the 2 points stats using PCA
• Classification
  • Train test splits
  • Data pipelines
  • Cross-validation
• Speed things up using Dask

4. Learn a Cahn-Hilliard simulation using regression

• Generate sample Cahn-Hilliard data
• Discretize the data using a continuous microstructure function
• Regression analysis to avoid repeating the simulations
  • generating the MKS "kernel" or "coefficients" using MKS "localization"
  • scaling up the "coefficients"
• Compare 2 microstructure functions
  • Train test splits
  • Data pipelines
  • Cross-validation
• Speed things up with Dask