Topics
24 topics across 76 articles, in 5 groups that follow the order of the work.
Inside a group the deepest topics come first, and every bar is drawn on the same scale. 5 topics hold ten articles or more; 11 hold fewer than five.
Foundations and maths
What each method actually does, and the statistics and probability underneath it. This is where the archive is deepest.
- Foundations
33 articles
The core ideas of data science and AI explained from scratch — what each method actually does, before any library gets involved.
Longest Monte Carlo methods: answering hard questions by rolling dice · 11 min - Statistics
14 articles
The inference layer: what your numbers support, what they do not, and the tests people reach for too quickly.
Longest Correlation is not causation, but · 8 min - Probability
7 articles
Reasoning about uncertainty, from Bayes and the central limit theorem to the puzzles that break everyone’s intuition the first time.
- Information Theory
3 articles
Information as surprise: entropy, bits, and the direct line from there to the loss functions used every day.
Data and SQL
Getting the rows out, then making them trustworthy enough to model. The part of the job that takes the most time.
- Data Cleaning
7 articles
The bulk of the job. Making messy data trustworthy before any model touches it.
- Feature Engineering
5 articles
Turning raw columns into inputs a model can use, without building something that breaks in production.
- SQL
4 articles
Getting the data out and computing where it is cheapest to compute.
Models and methods
The model families themselves, from a linear fit to a transformer, and what each one is worth on real data.
- Machine Learning
22 articles
Models that learn from examples — how the common families work, and when a simpler one would have been better.
Longest DBSCAN, and when density beats centroids · 11 min - Deep Learning
10 articles
Neural networks and what they buy you: real power on unstructured data, and a poor trade on most tables.
Longest How diffusion models make images · 8 min - LLMs
6 articles
Large language models: what next-token prediction produces, why it fails the way it does, and how to build on it honestly.
- Clustering
3 articles
Finding structure without labels, and staying honest about how much of it is real.
- Computer Vision
3 articles
How machines process images: convolution, generation, and the limits of both.
- Neural Networks
3 articles
The architecture itself — weights, layers, activations, and the search that sets them.
- NLP
3 articles
Working with text — from TF-IDF, which still earns its place, through to embeddings.
- Reinforcement Learning
3 articles
Learning from reward rather than labels. Powerful in games, rare in normal data science work, and this explains why.
- Time Series
3 articles
Data with an order, where the usual habits of splitting and shuffling quietly break.
Evaluation and evidence
Whether a number means anything: the metric that matches the decision, and the test that supports the claim.
- Evaluation
14 articles
Measuring a model honestly — the metric that matches the decision, the leaks that flatter it, the score you can defend.
Longest Data leakage is why your model looks too good · 10 min - Experimentation
5 articles
A/B tests and the discipline around them: pre-specify, size it, resist peeking.
- Causal Inference
4 articles
Separating what causes what from what merely moves together — and what it takes to claim the difference.
- Interpretability
3 articles
Understanding why a model predicted what it did, and being honest about how far that understanding goes.
Working practice
Everything around the model — framing the question, the code, production, and explaining the result to people who did not build it.
- Python
8 articles
Practical code for the work — pandas, scikit-learn, numpy, used the way practitioners actually use them.
- Communication
6 articles
Presenting results to people who do not care about your architecture, and should not have to.
- Production
6 articles
Getting a model out of the notebook and keeping it working once real traffic hits it.
- Framing
3 articles
Working out what the question actually is before answering it — the decision underneath the request, the unit it should be measured in, and when the honest answer is that it is not a data problem.