Machine Learning Foundations
Phase 5 of the quantitative learning path. You've learned factors and traditional quant. Now learn the ML machinery. and the brutal discipline required to use it without fooling yourself.
8 articles ~1.3 h
Concepts
- What Is Machine Learning, Really? . Pattern recognition. The three paradigms (supervised, unsupervised, reinforcement) and why ML is mostly data work, not algorithm work.
- Supervised Learning: Regression and Classification . Two questions: how much (regression) or which one (classification). Everything else is decoration.
- Overfitting and Regularization: The Central Problem of ML . The central problem of ML. A model that fits training data perfectly is broken. L1/L2, dropout, early stopping, and the five quant-specific overfitting traps.
- Cross-Validation and Honest Evaluation . Where strategies go to die. Random splits lie in time series. Walk-forward, purged CV, and the Deflated Sharpe Ratio.
Core Algorithms
- Linear and Logistic Regression . The boring baselines. Start here, end here if your problem is well-behaved. The first principle of ML research.
- Decision Trees and Random Forests . Flowcharts that learn from data. Random forests: a thousand mediocre opinions beat one confident one.
- Gradient Boosting: XGBoost and LightGBM . The most successful algorithm in tabular ML. XGBoost and LightGBM. If you only learn one model, learn this.
- Neural Networks: The Honest Version . Flexible function approximators. They dominate images and text. In quant, they're increasingly important, but GBM is usually better for tabular.