// Stage 2 · Classical Machine Learning
The ML Workflow: Data, Loss & Generalization
Train/validation/test splits, overfitting, bias–variance, cross-validation and metrics.
about 10 hours3 algorithms1 paperslesson text ready
01Goals
- Know what training, validation and test sets are for
- Spot over- and underfitting and use regularisation
- Choose accuracy, precision, recall, AUC and other metrics
Do these first: Calculus & Optimization · Probability, Statistics & Information Theory
02Read on this site
03Algorithms to master
Each card has the original paper, prerequisites and resources.
04Must-read papers
05Courses & videos
- cs229.stanford.edu ↗CoursesEN
- developers.google.com ↗CoursesMULTI
- statlearning.com ↗BooksEN
- coursera.org ↗CoursesEN
06Hands-on: project, code & exercises
Free, official or author-published pages. Most run in the browser or on a free Colab GPU.
PROJECT
Build a logistic-regression baseline on a Kaggle tabular dataset, run 5-fold cross-validation and plot learning curves.
- 1Kaggle Learn：Intro to Machine Learning ↗Exercises
Complete the online exercises from loading data to modelling and validation.
- 2
- 3Kaggle Learn：Intermediate Machine Learning ↗Exercises
Practise missing values, categorical features, pipelines and leakage.
07Self-test
Answer in your head, then open to check.
Why can’t you tune on the test set?
It leaks test information into model selection and makes the evaluation look too good.
Why is accuracy misleading on very imbalanced classes?
Predicting the majority class every time already scores high.
What does L2 regularisation do?
It penalises large weights and reduces overfitting.
On these routes: Algorithm Researchers · Domain Scientists