Article library

Explore by topic

Choose a subject and follow the ideas in the order that feels useful to you.

Topic

Decision trees

How trees split, predict, and become stronger together.

5 articles
  1. Decision-tree architectures

    Beyond CART: Decision Trees Don't Have to Look Like CART

    Identify the four architectural assumptions behind CART, then reason about what oblique, soft, oblivious, and model trees gain and give up by changing them.

  2. Gradient boosted decision trees

    Gradient Boosting: A Quick Deep Dive

    Start from one constant prediction, fit small trees to the corrections that remain, and discover how a chosen loss turns boosting into gradient descent in function space.

  3. Bagging and model variance

    Why averaging trees makes predictions more stable

    Keep one input fixed, watch tree predictions move as the data changes, and use averaging to narrow their distribution.

  4. Regression trees

    How decision trees learn to predict numbers

    Keep the same propose–judge–repeat procedure, replace class impurity with squared error, and let each leaf predict its mean.

  5. Decision trees

    How to invent the decision tree algorithm from scratch

    Build a classification tree by hand from ten seeds, then automate your own judgment with midpoint thresholds and the Gini score.

Topic

Reinforcement learning

How agents learn from rewards, experience, and state.

3 articles
  1. The Markov principle and MDP state designLatest

    The Markov Principle

    Compare Markovian and non-Markovian state representations in Pong, a door-and-key gridworld, and an ordered-checkpoint task.

  2. Sample-based reinforcement learning

    A Sample-Based View of Reinforcement Learning

    Follow experience through an MDP, then compare how reinforcement learning algorithms collect samples, update learned quantities, and represent what they learn.

  3. Reward hacking

    Reward Hacking in Reinforcement Learning

    Three small MDPs show how an agent can exploit a missing restriction, manipulate a measurement process, or optimize an incomplete objective.

Topic

Time series

How models find useful structure across time.

1 article
  1. ACF and PACF

    How the Past Echoes: Understanding ACF and PACF

    Begin with ordinary correlation, shift a series through time, and see how lagged relationships become an autocorrelation function.

Topic

Foundations and evaluation

Core ideas for understanding and judging machine-learning models.

1 article
  1. Accuracy, precision, and recall

    Is My Model Good?

    Compare what accuracy, precision, and recall reveal about the same predictions, then choose the metric that matches the mistake that matters.