Overview
Objectives
- Get an overview of (geospatial) machine learning.
- Understand framework and key methods of supervised learning.
- Can apply supervised learning to real-world problems with various data types (tabular, image, network)
Why study this module?
- AI can make mistakes; understanding how it works is crucial.
- Achieving a balance between predictive accuracy and interpretability is important.
- Spatial is special: spatial data have unique characteristics that require special treatment.
Prerequisites
- Can understand/follow Python code and run Jupyter notebooks.
- Basic knowledge of statistics (mean, variance, correlation, probability) and linear regression.
Course Structure
- Lectures: 10 weeks (Wednesdays 9:00–10:30).
- Tutorials: 10 weeks (Wednesdays 10:30–12:00).
Platforms
- Email for important notices and private questions.
- Github & website for lecture notes and notebooks.
- Moodle for lectures recording and assessments.
- Slack for public questions.
Weekly Schedule
| Session | Topic | Suggested reading |
|---|---|---|
| 1 | Intro machine learning | IMLP Ch 1 |
| 2 | Supervised learning metrics | IMLP 5.3, APM Ch 16 |
| 3 | Supervised learning workflow | IMLP Ch2.1-2.3.2, APM Ch 4-4.3, IMLP Ch 5.1, 5.2, APM Ch 4.4-4.8 |
| 4 | Tree-based methods | IMLP 2.3.5, 2.3.6, APM Ch 14.1-14.4 |
| 5 | Neural networks | IMLP Ch 2.3.8, DL Ch 6, Ch 7.8 |
| 6 | Graph neural networks | tkipf.github.io |
| 7 | Model interpretation & feature selection | IML, LIML |
| 8 | Imbalanced data | APM Ch16, SMOTE, Easy Ensembles |
| 9 | Machine learning operations | Madewithml.com |
| 10 | Testing ML systems | Madewithml.com |
References
- (COMS) COMS W4995 Applied Machine Learning Spring 2020 link
- (IMLP) Mueller, Guido - Introduction to machine learning with python. link, free for UCL users
- (APM) Kuhn, Johnson - Applied predictive modeling link
- Goodfellow, Bengio, Courville - Deep Learning. link
- (PDSH) Python Data Science Handbook by Jake VanderPlas link
- (IML) Interpretable Machine Learning by Christoph Molnar link
- (LIML) Limitations of Interpretable Machine Learning link
Attendance Recording
- 70% attendance required for student visa holders.
- Please attend all lectures and workshops.
- Contact module lead & bartlett.pg-casa@ucl.ac.uk, if you can’t attend.
Assessment
Assessment
- Python notebook (summative): 100%
- Weekly quiz (formative)
UCL Assessment Policy
- All submissions via Moodle, not emails.
- Late penalties: Up to 48h (-10 points); up to 7 days (capped at 50); over 7 days (scores 0).
- DAP or Extenuating circumstances: to submit on Portico.
- Avoid plagiarism and unverified references
Moodle Feedback
Please provide anonymous feedback on Moodle for every week.
Github Feedback
You can also give feedback on Github issues.