AutomaticMALA

AutoMALA

An automatic version of the Metropolis-adjusted Langevin algorithm (MALA).

See Biron-Lattes, Surjanovic & Syed et al. (2023) for details.

Fields

• ϵ_init::Any: initial step size to use

• num_unadjusted::Int64: number of unadjusted steps to take

round_based_adaptation([rng, ]model, sample::AutoMAL; num_rounds=10, kwargs...)

Run adaptation for a number of rounds, each time doubling the number of iterations.

Arguments

• rng::Random.AbstractRNG: Random number generator.
• model: The model.
• sample::AutoMALA: The sampler.

Keyword Arguments

• num_rounds::Int=10: The number of rounds.
• kwargs...: Additional keyword arguments passed to sample.

Returns

• sampler::AutoMALA: Sampler containing the adapted parameters.
• initial_params: Resulting parameters after adaptation.