Machine Learning from Scratch
Learn machine learning by building it — every algorithm implemented by hand and run in your browser
An interactive machine learning course built for the AI era. Because an AI can already write model.fit() for you, this course teaches the durable skill instead: you implement gradient descent, logistic regression, decision trees, and k-means yourself in Python that runs live in your browser, build the evaluation tools that catch expensive mistakes (the accuracy trap, leakage, drift, Simpson's paradox), and then meet scikit-learn with an understanding of exactly what it does. 10 modules, 54 lessons, roughly 16 hours, with 380 exercises and quiz questions that actually execute and grade in the browser.
- ✓ Algorithms from Scratch
- ✓ Honest Evaluation
- ✓ The Traps
- ✓ The Professional Stack
This course includes:
- 📹 16 hours of video content
- 📝 10 comprehensive modules
- 💻 Hands-on coding projects
- 🏆 Certificate of completion
- 🔄 Full access during your selected plan
Course Overview
What You'll Learn
Master these essential skills that employers are actively seeking
Algorithms from Scratch
Gradient descent, logistic regression, decision trees, k-means — implemented by hand, not imported
Honest Evaluation
Confusion matrices, ROC/PR curves, calibration, and thresholds under asymmetric cost
The Traps
Data leakage, drift, spurious correlation, Simpson's paradox, imbalance — reproduced and defended against
The Professional Stack
scikit-learn pipelines used correctly, verified at build time, plus classical-vs-LLM judgment
How It Works
Build the Math
Implement each algorithm from scratch in stdlib Python, live in your browser
Run It for Real
Every exercise executes — predictions checked against real output, not multiple-choice hand-waving
Meet the Traps
Reproduce leakage, drift, and the accuracy trap so you recognise them in the wild
Graduate to scikit-learn
Map what you built onto the professional stack and ship an honest end-to-end pipeline
Tips for Success
Type the algorithms out yourself — the skill is in the implementing, not the reading
Watch the loss numbers as you change the learning rate; diverging loss teaches more than any paragraph
Module 9 (How You Get Deceived) is the highest-value module for real work — do not skip it
When you reach scikit-learn in Module 10, map each API call back to the loop you built by hand
Frequently Asked Questions
Comfortable arithmetic and a willingness to read a formula slowly. Every equation is implemented as short Python code you can run, which is gentler than notation.
Basic Python (variables, functions, loops) is assumed. Our Python course covers everything needed.
Yes — lessons execute real Python in your browser via WebAssembly. The scikit-learn lessons are shown as verified output rather than run in-browser, and we say so on the lesson.
It builds the foundations deep learning rests on (gradient descent, loss surfaces, overfitting) and covers where classical ML fits next to large language models. It does not train neural networks.
Related Courses
Ready to Master Machine?
Learn by building and running real code, one lesson at a time