Quap Lab

Write a Bayesian model as a formula list, paste your data as CSV or JSON, and fit it. quap puts a Gaussian at the posterior mode in milliseconds. NUTS samples the real shape of the posterior in seconds.

The approach is not mine. The formula-list notation, quadratic approximation as the way in, and the examples this is built around all come from Statistical Rethinking by Richard McElreath. This is an independent browser implementation of ideas that book teaches, written because working through them without an R install seemed useful. If the subject interests you, read the book.

• A formula list with ~ for stochastic and <- for deterministic relations, twelve distributions.
• Posterior summaries with compatibility intervals, plus a grid cross-check that says when the approximation is off.
• Prior predictive and posterior predictive simulation, so you can see what a prior actually believes.
• WAIC and PSIS-LOO with Pareto k diagnostics for comparing models.