Machine Learning
Machine learning from its exact arithmetic core: small rational data, validated deterministic diagrams, and clear boundaries around what is computed versus what is not claimed.
Books
- Linear Regression (Least Squares)
- Ridge Regression, Exactly
- Lasso, Exactly
- Why Toys Can't Scale
- Feature Scaling, Exactly
- Categorical Features, Exactly
- Missing Values, Exactly
- Train/Test Split, Exactly
- Feature Matrix, Exactly
- Class Imbalance, Exactly
- Train-Only Preprocessing, Exactly
- Decision Thresholds, Exactly
- Precision and Recall, Exactly
- F1 Score, Exactly
- Calibration Bins, Exactly
- Logistic Scoring
- Decision Trees (Gini)
- k-NN and Boundaries
- The Perceptron
- Max-Margin (SVM)
- Naive Bayes from Counts
- k-Means, One Honest Step
- A Neural Net You Can Compute by Hand
- Backpropagation, Exactly
- Gradient Descent, One Step
- Stochastic Gradient Descent, Exactly
- Tokenization (BPE)
- Embeddings
- Attention, Structurally
- The Transformer Block
- Decoding
- A Transformer, End to End
- Running a Real Model
- Evaluation, Honestly
- Cross-Entropy, Honestly
- Cross-Entropy, Exactly
- Convolution, Exactly