Advanced Topics in Chemical Physics: Introduction to Machine Learning

This repository contains five lectures and three workshop sessions on introducing machine learning concepts in the advanced physical chemistry module at UoE.

Authors

Dr Antonia Mey -- antonia.mey@ed.ac.uk.
Dr Matteo Degiacomi -- matteo.t.degiacomi@durham.ac.uk

Jasmin Güven, Rohan Gorantla, Ryan Zhu

Demonstrators 24-25:

Ryan Zhu and Dominic Philips

Workshop Notebooks

Workshop	Materials
Workshop 01: Clustering and Regression
1. Clustering
2. Regression
3. Application
Workshop 02: Dimensionality Reduction and Classification
1. Dimensionality reduction
2. Classification
Workshop 03: Deep Learning for Solubility Classification
1. Intro to Pytorch
2. Solubility classification

Local installation

1. Install anaconda.

2. Create a new environment:

   conda create -n ml_chem

3. Activate the environment:

   conda activate ml_chem

4. Install mamba to make the installation of packages faster.

   conda install -c conda-forge mamba

5. Install all the required packages with mamba:

   mamba install -c conda-forge scikit-learn matplotlib pandas

For Unit_03 you will also need to install

mamba install -c conda-forge rdkit seaborn

and

mamba install pytorch torchvision torchinfo -c pytorch

Project

Release: week 4 Report Deadline: TBC
Weight: 20%

Summary of Lectures

Lecture 1:

• What is machine learning?
• Examples of machine learning (in Chemistry)
• Linear Regressions
• Introduction to unsupervised learning:
 - Clustering (k-means and others)

Lecture 2:

• Chemistry data is high dimensional
• Dimensionality reduction:
  • Principle component analysis (PCA)
  • Time lagged component analysis (tICA)
  • t-stochastic neighbour embedding (t-SNE)

Lecture 3:

• Classification problems
• Classifications in practice:
  • Random Forests
  • Support vector machine

Lecture 4:

• Shallow Learning
• Deep Learning part I
  • Multilayer perceptron


Lecture 5:

• Deep Learning Part II
  • Transformers
  • Graph Neural Networks



Learning Outcomes

• Understand the main pillars of machine learning
• Know about different clustering techniques as part of unsupervised learning
• Be able to use common nomenclature used in machine learning
• Use Principle component analysis to reduce the dimensions of a data set
• Understand how a regression problem can be cast as a machine learning problem
• Be aware of how random forests and multilayer perceptrons can be used in a classification problem

Additional Resources

A handout with additional resources can be found here.