Metropolis-Hastings Method
In Importance Sampling and Rejection Sampling, the proposal
distribution
Method
- Evaluate
for any . - A tentative new state
is generated from the proposal density . - Compute
- If
, accept new state and set , else set
Pros and Cons
- Will still give answers in high-dimensional settings
- Lengthy simulations may be needed for convergence, because of the quadratic dependence on the lengthscale-ratio. A random walk is extremely slow, and should try to be suppressed.
Suppressing Random Walks
Hamiltonian Monte-Carlo methods make use of gradient information to reduce random-walk behaviour.