-
Topics Everyone Is Talking About No426
We should be more ambitious with the software we build • A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis • I refuse to accept a censored Internet. Join Me at FOSDEM 27 • Dolly Parton’s Imagination Library • OpenAI and Anthropic cut their LLM prices by up to 80% within a few months.…
-
From EDA to features: encoding, scaling, and the leakage traps in between
A machine learning model can achieve a perfect validation score and still be completely useless in production.
-
Dimensionality reduction for seeing: PCA, t-SNE, and UMAP compared honestly
A dataset with 64 features cannot be plotted directly on a two-dimensional screen. But a machine learning engineer can transform those 64 features into two coordinates and produce a surprisingly beaut
-
Simpson’s paradox: when your aggregate answer is exactly wrong
Imagine you are a data analyst at an e-commerce company. Your team uses two shipping carriers, ExpressSwift and BudgetFlow, and management wants to know which one delivers packages on time more reliab
-
Regularizing a boosted model: reg_lambda, reg_alpha, gamma, and the constraints that matter
XGBoost does not regularize a tree in quite the same way that ridge regression or Lasso regularizes a linear model. A standard tree booster does not have one coefficient per input feature. Instead, it
-
Detecting data drift in production with the K–S test and PSI
A machine-learning model can be perfectly healthy from an infrastructure perspective and still become increasingly wrong.
-
Chi-squared tests for categorical data: independence and goodness-of-fit
Many important questions in data analysis involve categories rather than numerical measurements:
-
Regularization in practice: telling underfitting from overfitting and choosing the strength
A machine learning model can fail in two opposite ways.
-
Bootstrap and permutation tests: confidence intervals and p-values without a formula
You have two versions of an API endpoint. The new version appears faster, but its latency distribution is skewed, a handful of requests take much longer than the rest, and the median matters more to y