1. Home
  2. Courses
  3. The machine learning path

A suggested order

The machine learning path

15 stages from your first line of Python to machine learning in production, with each maths and statistics lesson placed just before the model that uses it. Follow it in order, or start at the stage you need. Every course can also be taken on its own.

Stages
15
Checkpoints
14
Lessons
172 open of 209
Courses
Python, Data Tools, Maths, Statistics, Machine Learning
Start at Stage 0 Find your starting stage

Where to start

Start at the stage that fits you

Where to start on the machine learning path, and what to skip
You areStart atSkip
New to programmingStage 0Nothing
Comfortable with basic Python (variables, loops, functions, lists)Stage 1Stage 0. Take Python Module 6 before Stage 4 if you have not used classes
A programmer who knows NumPy, pandas and school mathsStage 2Stages 0 and 1. Take any foundation lesson a lesson lists that you cannot answer
Already know some machine learningThe first checkpoint you cannot yet doEvery stage whose checkpoint you can already do

The stages

Every stage, with its checkpoint

The checkpoint is what you should be able to do before moving on. If you can already do it, skip the stage.

  1. Stage 0

    Python from zero

    CheckpointRead a CSV file by hand, count how often each word appears using a dictionary, and print the ten most common words. Write a pytest test for the counting function, and make Ruff and mypy report nothing on it.

    Skip it if you can already write a function that loops over a list.

    27 lessons

    Python

    Start with Python 1.1 →

  2. Stage 1

    Your tools, and reading maths

    CheckpointLoad a table with pandas, compute the average of one column per group, and plot it. Read ŷᵢ = Σⱼ wⱼ xᵢⱼ aloud in words.
    24 lessons

    Data Tools

    Statistics

    Maths

    Python

    Start with Tools 0.1 →

  3. Stage 2

    What machine learning is

    CheckpointGiven ten real problems, say which are regression, which are classification and which are unsupervised. Name the features and label of each.
    4 lessons

    Machine Learning

    Start with ML 1.1 →

  4. Stage 3

    Understanding data

    CheckpointProduce a written data report on a dataset you have never seen, using the ML 2.2 checklist. Reproduce one tempting claim from that dataset, check its denominator and plot choices, and rewrite it with its limitations.
    18 lessons

    Data Tools

    Statistics

    Machine Learning

    Start with Tools 2.4 →

  5. Stage 4

    Your first model: linear regression

    CheckpointImplement linear regression with gradient descent from scratch in NumPy, and match scikit-learn's coefficients (the ML 3.3 assignment). Explain why a coefficient changes when another feature is added, and why correlated features make coefficients unstable while the predictions hold.
    23 lessons

    Maths

    Machine Learning

    Python

    Data Tools

    Start with Maths 2.2 →

  6. Stage 5

    Classification, and knowing whether a model works

    CheckpointTune K for k-NN with cross-validation, touch the test set once, and report precision, recall and AUC. Choose a ridge penalty by cross-validation and explain what lasso does that ridge does not. Explain why the training score is not evidence.
    20 lessons

    Maths

    Machine Learning

    Data Tools

    Statistics

    Start with Maths 2.3 →

  7. Stage 6

    Linear classifiers

    CheckpointCalculate the log loss of individual predictions and explain why a confidently wrong answer costs more; train a logistic regression and explain one weight in plain words, including when you are not allowed to read it as importance. Deriving log loss algebraically is optional.
    14 lessons

    Maths

    Machine Learning

    Statistics

    Start with Maths 2.4 →

  8. Stage 7

    Probabilistic models and text

    CheckpointBuild a spam or sentiment classifier with a bag of words and Naive Bayes, compare it with TF-IDF and logistic regression by a cross-validated score, and explain why Naive Bayes adds log probabilities instead of multiplying probabilities.
    5 lessons

    Statistics

    Maths

    Machine Learning

    Start with Stats 4.3 →

  9. Stage 8

    Margins, kernels, and the first full project

    CheckpointComplete the ML 7.9 assignments, at least the easy and medium tiers, and present your comparison table.
    4 lessons

    Machine Learning

    Maths

    Start with ML 7.6 →

  10. Stage 9

    Preparing real data

    CheckpointTurn a messy table with missing values and categories into one Pipeline that cross-validates without leaking, and complete the ML 8.7 assignments.
    9 lessons

    Statistics

    Data Tools

    Machine Learning

    Maths

    Start with Stats 4.5 →

  11. Stage 10

    Trees, ensembles, and explaining models

    CheckpointBeat your logistic regression with gradient boosting on a tabular problem, choose the threshold from the costs, and explain the final model's biggest drivers.
    10 lessons · 7 open now

    Statistics

    Machine Learning

    Start with Stats 5.3 →

  12. Stage 11

    Learning without labels

    CheckpointReduce a digits dataset with PCA, cluster it, and judge the clusters with and without the labels.
    14 lessons · 4 open now

    Maths

    Data Tools

    Machine Learning

    • ML 11.1Principal Component AnalysisComing soon
    • ML 11.2PCA: Why the EigenvectorsComing soonsecond pass
    • ML 11.3t-SNEComing soon
    • ML 12.1K-MeansComing soon
    • ML 12.2Hierarchical ClusteringComing soon
    • ML 12.3DBSCANComing soon
    • ML 12.4Gaussian Mixture Models and EMComing soon
    • ML 12.5Evaluating ClustersComing soon
    • ML 13.1Local Outlier FactorComing soon
    • ML 13.2Isolation ForestComing soon
    Start with Maths 3.5 →

  13. Stage 12

    Recommendations and time

    CheckpointForecast a series with walk-forward validation and beat the seasonal-naive baseline.
    7 lessons · 2 open now

    Machine Learning

    • ML 14.1Collaborative FilteringComing soon
    • ML 14.2Matrix FactorisationComing soon
    • ML 15.1Time Series BasicsComing soon
    • ML 15.2Forecasting as RegressionComing soon
    • ML 15.3Smoothing and ARIMAComing soon

    Data Tools

    Statistics

    Start with Tools 2.7 →

  14. Stage 13

    Machine learning in the real world

    CheckpointTake your Stage 10 model and calibrate it, put a confidence interval on its score, and save it as a pipeline that could be served. Then write down how you would monitor it and what would make you retrain.
    13 lessons · 4 open now

    Statistics

    Machine Learning

    • ML 17.1Probability CalibrationComing soon
    • ML 17.2Is Model B Really Better?Coming soon
    • ML 17.3From Notebook to ProductionComing soon
    • ML 17.4Rolling Out a New Model: Shadow, Canary and A/BComing soon
    • ML 17.5Monitoring, Drift and RetrainingComing soon
    • ML 17.6Reproducibility and Experiment TrackingComing soon
    • ML 17.7FairnessComing soon
    • ML 17.9Prediction Intervals and Sets: Conformal PredictionComing soon
    • ML 17.10Choosing an AlgorithmComing soon
    Start with Stats 6.1 →

  15. Stage 14Optional

    Going further

    18 lessons · 8 open now

    Statistics

    Machine Learning

    • ML 4.10k-NN at Scale: kd-Trees and HashingComing soon
    • ML 7.5Multi-Label ClassificationComing soon
    • ML 9.5AdaBoostComing soon
    • ML 13.3One-Class SVMComing soon
    • ML 14.3Learning to RankComing soon
    • ML 17.8PrivacyComing soon
    • ML 16.1Semi-Supervised LearningComing soon
    • ML 16.2Active LearningComing soon
    • ML 18.1Case Study: Human Activity RecognitionComing soon
    • ML 18.2Case Study: Friend RecommendationComing soon
    Start with Stats 1.10 →

Why this order

Built so what you learn is used straight away

Each foundation, just before you need it

You learn the dot product just before k-nearest neighbours and Bayes' theorem just before Naive Bayes, so what you learn is used straight away.

Wide first, then deep

On the first pass, learn what each method does and when to use it. Lessons marked second pass go deeper; come back to them later.

A checkpoint at every stage

Each stage ends in something to do, not something to read. If you can already do it, skip the stage.

Do the notebook, don't just watch

Watching gives the feeling of understanding; the notebook tells you whether you have it.

Prefer to pick your own order? Browse the courses; each one lists what it needs first.