← ~/visualizations

bayesian-inference

Visualizes a full Bayes update over a parameter θ in [0,1]: a Beta prior p(θ), a Binomial likelihood p(x|θ) from animated data (n trials, k successes), and the resulting posterior p(θ|x). The animation phases highlight the conceptual pipeline (prior → likelihood → multiply → normalize by evidence), while the equation and evidence term are shown in a HUD box.

canvasclick to interact
t=0s

practical uses

  • 01.Updating a coin-bias estimate as flips arrive (A/B testing conversion rates)
  • 02.Bayesian parameter estimation and uncertainty quantification in ML models
  • 03.Online decision-making: continually revising beliefs as new measurements come in

technical notes

Renders prior/likelihood/posterior as blocky histograms on a θ-grid with green-on-black styling. Uses Beta-Binomial conjugacy for posterior (Beta(a+k,b+n−k)) and approximates evidence via discrete integration of prior·likelihood. Animation is time-phased over a 4s cycle with easing; data k updates discretely to show how posterior reacts.