Machine Learning — EGEI
Learn the foundations of machine learning while working the way data teams actually work today: side by side with an AI coding assistant. You will not just prompt it — you will read, test and defend every line it writes.
One case runs through the whole course: predicting household poverty with CASEN. Every session, with its notebook, is in Classes; deadlines and rubric are in Project; the formal syllabus is in Program.
What this course is about
Machine learning has two halves. One is statistical: what a model is, what it can and cannot learn from data, and how to tell a real signal from noise. The other is craft: turning a business question into data, code and a decision someone can act on.
This year the craft half changes. Writing code is no longer the bottleneck — reviewing it is. So every session pairs a modelling topic with the corresponding assistant skill: how to ask for the right thing, how to check the answer, and how to recognise when the assistant is confidently wrong.
Modelling foundations
Regression, classification, resampling, trees, regularization and clustering, all on one real
policy problem, with emphasis on why a model works, not only on calling .fit().
Assistant-driven workflow
Use Copilot, Claude, ChatGPT or Cursor as a pair programmer: scaffolding, refactoring, debugging and explaining code — under a verification discipline you can defend.
One real project
Everything is assessed through one end-to-end project on real data, in two deliveries, with code, a written report and a live presentation.
Learning outcomes
By the end of the course you will be able to:
- Frame a business question as a supervised learning problem, and say what a good answer looks like.
- Prepare, explore and visualize a real dataset in Python, reproducibly.
- Choose, fit and tune regression and classification models, and justify the choice.
- Evaluate models honestly: train/test discipline, cross-validation, uncertainty and the right metric.
- Work with an AI coding assistant productively — and verify its output rather than trusting it.
- Communicate results and limitations to a non-technical audience.
What you need
| You need | Details |
|---|---|
| Python environment | Local install (Anaconda/uv/venv) or Google Colab — both work |
| Code assistant | At least one of: GitHub Copilot, Claude, ChatGPT, Gemini, Cursor (free tiers are enough) |
| Background | Basic programming and basic statistics; no prior ML required |
See the Code Assistants page for setup, prompting patterns, the verification checklist and the course policy on assistant use.
Instructor
Sebastián Azócar
Data Scientist @ LATAM Airlines Group · Lecturer
Mathematical Engineer (USACH) with an MSc in Data Science (Universidad del Desarrollo). Over five years applying machine learning in industry — Caja Los Andes, Walmart Chile and currently LATAM Airlines Group — and more than a decade teaching data science and statistics.
Material and student projects from previous editions are in the Archive.