Machine Learning
Four levels, twelve lessons, from a first model to one you can defend
Machine learning is fitting a function to examples. You show a program thousands of cases where you know the answer, it finds a rule that reproduces them, and you then apply that rule to cases where you do not know the answer. There is no understanding involved, and treating it as if there were is the source of most disappointment with it.
Overview
It is the point in this track where the earlier courses pay off. Everything here assumes you can load and clean data in pandas and that you know why a difference might be noise.
Introduction
It assumes Python and pandas - the first course in this track is enough - and no mathematics beyond arithmetic. Where a method needs an idea from statistics, the idea is explained in words where it is used.
Every level is three lessons and a Level Check quiz, with an Expert Exam at the end. Examples use scikit-learn with small built-in datasets, so every one runs without a download.
By the end you will be able to frame a problem as a learning task, build a model that does not fool you, measure it with a metric that matches the cost of being wrong, and say honestly what it will and will not do once it is live.
What you will be able to do
- Frame a business problem as a supervised learning task, or rule out that it is one
- Split data so your evaluation means something, including for time series
- Recognise overfitting and leakage before they reach a slide
- Choose a metric that matches the real cost of each kind of error
- Pick between linear models, forests and boosting for a reason you can state
- Put a model behind a pipeline and monitor it for drift once it is live
What each level covers
4 levels · 16 lessons
-
Level 1 - Basic
What a model actually is, and a first one built properly. The vocabulary comes first because it is used imprecisely everywhere: features, labels, supervised and unsupervised, classification and regression. Then the train/test split, which is the one idea that makes evaluation meaningful, and a complete first model with its score interpreted rather than just reported.
4 lessons 40 min of reading
Start with What a Model Is, and What It Is Not → -
Level 2 - Intermediate
Why a first model is usually wrong, and how to find out before anybody else does. Overfitting and the cross-validation that measures it, feature engineering and the leakage that masquerades as brilliance, and the evaluation metrics that matter once accuracy has been exposed as a poor summary.
4 lessons 46 min of reading
Start with Overfitting, Underfitting and Cross-Validation → -
Level 3 - Advanced
The models worth knowing, and when each is the right answer. Linear and logistic regression first, because they are interpretable and frequently sufficient; then trees, forests and gradient boosting, which win most competitions on tabular data; then the unsupervised pair - clustering and dimensionality reduction - with an honest account of how to tell a real cluster from an arbitrary one.
4 lessons 46 min of reading
Start with Linear and Logistic Regression → -
Level 4 - Expert
Everything between a good notebook and a model that keeps working. Pipelines and tuning done so the result is reproducible rather than lucky; drift, which is the reason a model that was fine in March is wrong in September; and the obligations that arrive the moment a model makes a decision about a person.
4 lessons 58 min of reading
Start with Pipelines, Tuning and Reproducibility →
Guides and articles
Longer reading about this subject, written alongside the course.
Before you start
- Python and pandas - the first course in this track is enough
- No mathematics beyond arithmetic
Common questions
- What does the Machine Learning course cover?
- It is 4 levels and 16 lessons: Level 1 - Basic; Level 2 - Intermediate; Level 3 - Advanced; Level 4 - Expert. Each level ends with a quiz, and the last one is an exam over the whole course.
- How long does it take?
- About 190 minutes of reading and exercises - roughly 3.2 hours. It is self-paced with no cohort and no start date, so you can leave and come back and the site remembers where you were.
- What do I need to know first?
- Python and pandas - the first course in this track is enough. No mathematics beyond arithmetic.
- Is it free?
- Yes. Every lesson, example and quiz in this course is free to read, and you do not need an account to read them. Signing in only adds progress tracking and quiz results.
- What will I be able to do afterwards?
- Frame a business problem as a supervised learning task, or rule out that it is one. Split data so your evaluation means something, including for time series. Recognise overfitting and leakage before they reach a slide. Choose a metric that matches the real cost of each kind of error. Pick between linear models, forests and boosting for a reason you can state. Put a model behind a pipeline and monitor it for drift once it is live.