Hamropedia

Phase 2 of 7 · 17 topics

Classical ML, built by you

Build every classic algorithm yourself in NumPy. After this phase, a library is only a shortcut for something you already understand.

Classical ML is a toolbox of different tools for one job: learning a pattern from examples. Build each tool once by hand, and you will know which one to pick, and why.

0 of 17 done

Core resources for this phase

Topics, in order

  1. 2.1 The ML setup: features, labels, loss, train/validation/test

  2. 2.2 Linear regression: normal equation and gradient descent

  3. 2.3 Logistic regression and the sigmoid

  4. 2.4 Regularisation: L1 and L2

  5. 2.5 Bias–variance trade-off and overfitting

  6. 2.6 Metrics: precision, recall, F1, ROC-AUC, confusion matrix

  7. 2.7 Cross-validation and data leakage

  8. 2.8 Feature scaling and encoding

  9. 2.9 k-nearest neighbours

  10. 2.10 Decision trees: entropy and Gini

  11. 2.11 Ensembles: bagging, random forests, gradient boosting

  12. 2.12 Support vector machines: margin and hinge loss

  13. 2.13 Naive Bayes

  14. 2.14 k-means clustering

  15. 2.15 PCA in practice

  16. 2.16 Gaussian mixture models and EM

  17. 2.17 Generalisation: VC dimension and double descent

Go deeper (optional)