A set of digital notes made from 53 pages of handwritten Machine Learning notes. The handwriting has been typeset, the equations rewritten in LaTeX, and the sketches redrawn as diagrams placed right next to the explanation they support.
Who is it for? University students learning Machine Learning who want short, structured notes for learning and revision.
The notes follow the topics of the source and nothing else. Wherever the handwriting was unclear, or looked like a slip, it is flagged instead of silently fixed — see TRANSCRIPTION_NOTES.md. Text added to help understanding is always marked Additional Explanation.
- Start with supervised learning — everything else builds on its cost functions and gradient descent.
- Move to neural networks, which reuse those ideas.
- Learn how to evaluate and improve models (bias, variance, error analysis).
- Study decision trees and ensembles as the other major family of supervised models.
- Continue with unsupervised learning (clustering, anomaly detection).
- Finish with recommender systems and reinforcement learning.
Regression and classification: linear and logistic regression, gradient descent, feature scaling, overfitting and regularization.
- Linear Regression
- Linear Regression with Multiple Features
- Feature Scaling
- Gradient Descent in Practice
- Feature Engineering and Polynomial Regression
- Decision Boundary
- Logistic Loss, Cost and Gradient Descent
- Overfitting
- Regularization
Neural networks: layers, activation functions, softmax, vectorization, TensorFlow training steps, backpropagation, Adam.
- Neural Networks
- Layer Computation (Intuition)
- Activation Functions
- Multiclass Classification and Softmax
- For Loop versus Vectorization (Dense Layer)
- TensorFlow and Model Training Steps
- Backpropagation versus Gradient Descent
- Adam Algorithm and Types of Layers
Train/CV/test sets, model selection, bias and variance, learning curves, error analysis, metrics, transfer learning, deployment.
- Train / Cross-validation / Test Procedure
- Model Selection
- Bias and Variance
- Learning Curves and Debugging a Learning Algorithm
- Adding Data, Data Augmentation and Data Synthesis
- ML Development and Error Analysis
- Precision, Recall and the F1 Score
- Transfer Learning
- ML Project Cycle and Deployment
Decision trees, entropy, information gain, one-hot encoding, regression trees, random forests, XGBoost.
- Decision Trees
- Entropy as a Measure of Impurity
- Information Gain
- One-Hot Encoding
- Regression Trees
- Sampling with Replacement and the Random Forest Algorithm
- XGBoost (eXtreme Gradient Boosting)
- Decision Trees versus Neural Networks
K-means clustering, choosing K, the Gaussian distribution and anomaly detection.
- Clustering and K-Means
- K-Means Cost Function and Random Initialization
- Choosing the Value of K (Elbow Method)
- Gaussian (Normal) Distribution
- Anomaly Detection (Intuition) and Choosing Features
- Developing and Evaluating Anomaly Detection Systems
Collaborative and content-based filtering, retrieval and ranking, return, policy and the Q function.
- Recommender Systems: Notation
- Collaborative Filtering versus Content-Based Filtering
- Collaborative Filtering: Cost Function and Gradient Descent
- Collaborative Filtering with Binary Labels
- Mean Normalization
- Finding Related Items
- Content-Based Filtering with Deep Learning
- Retrieval and Ranking
- Reinforcement Learning: Return and Discount Factor
- Policy and the State-Action Value Function (Q)
- Digitised notes — every page of the handwritten PDF is accounted for.
- Mathematical equations — KaTeX/LaTeX equations that render on GitHub.
- Embedded visualisations — SVG diagrams and Matplotlib figures inside each topic page.
- Structured learning sequence — pages reordered by prerequisite, not by notebook order.
- Student-friendly explanations — every topic ends with a Detailed Explanation (intuition, worked examples, common mistakes), clearly labelled as an addition to the handwritten source.
- Source audit — TRANSCRIPTION_NOTES.md lists unclear handwriting, apparent errors and every added explanation.
- Each module folder has a
README.mdwith its topic list. - Each topic page ends with Related topics links to the previous, next and connected notes.
- Quick links: Linear regression · Neural networks · Bias and variance · Decision trees · K-means · Recommenders
The same Markdown files can be viewed as a dark-themed documentation site (Midnight Lavender palette, search, KaTeX):
pip install -r requirements.txt
bash scripts/serve_site.sh # opens http://127.0.0.1:8000pip install numpy matplotlib
python scripts/make_figures.py # assets/figures/*.png
python scripts/make_diagrams.py # assets/diagrams/*.svg
python scripts/build_indexes.py # module READMEs, this README, mkdocs.ymlColour palette: background #171821, cards #2C2B40, accent #A78BFA / #E89BB5, maths #EAC477, success #91C7A5.