liveBook · Manning

Proposal distribution

As explained in Grokking Bayes

A proposal distribution is a probability distribution from which candidate moves or states are generated during an iterative sampling algorithm.

A proposal distribution generates candidate moves for an iterative sampling algorithm. In Metropolis–Hastings, it suggests where the chain might go next, while an accept/reject rule decides whether the move becomes part of the sample trail. section 5.6

A useful proposal usually favors points near the current location, helping the chain remain in high-density regions, while sometimes suggesting farther points so it can explore elsewhere. Its suggestions are filtered against the target distribution. section 5.7

Metropolis–Hastings repeats a short cycle:

  • Start at a current point θ. section 5.6
  • Draw a candidate θ′ from a Normal proposal centered at θ, using a chosen standard deviation. section 5.6
  • Compare the unnormalized posterior densities with r = PDF(θ′) / PDF(θ). section 5.6
  • Accept θ′ with probability min(1, r); otherwise retain θ and try again. section 5.6

A small sampling walk

A Python implementation can set an initial value, choose a proposal standard deviation such as 0.05, generate Normal candidates, evaluate an unnormalized Beta(4, 3) posterior, and store either the candidate or the current value. Repeated accepted and repeated retained states form the Markov chain. section 5.6

How Grokking Bayes covers it

How deep it goes
Defines the proposal, specifies its Normal form, and walks through candidate generation and acceptance without proving convergence guarantees.
Examples
Uses a tourist-guide analogy, a Normal random walk, and a Beta(4, 3) posterior implemented with Python sampling code.
What it assumes
Assumes basic probability and posterior-density ideas while building MCMC intuition from an initial point and repeated candidate moves.
What you can do after
Apply it via Metropolis–Hastings to generate posterior samples from an unnormalized target density.
How it connects
Connects proposal distributions to MCMC, Metropolis–Hastings, posterior distributions, and accept/reject rules.

Also explained in

Compare how 3 books explain proposal distribution →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Grokking Bayes. The section links point to the parts of the book it draws on.