bristlecone
MetropolisWithinGibbs Module
Types and nested modules
| Type/Module | Description |
|
|
|
|
Log of the standard deviation (unitless) |
|
|
|
Functions and values
| Function or value |
Description
|
Full Usage:
core isAdaptive writeOut random domain f results batchLength batchNumber theta sigmas
Parameters:
bool
writeOut : LogEvent -> unit
random : Random
domain : Domain
f : Point -> TypedTensor<Scalar, MeasureProduct<-logL, MeasureOne>>
results : Solution list
batchLength : int<MeasureProduct<iteration, MeasureOne>>
batchNumber : int<MeasureProduct<batch, MeasureOne>>
theta : Point
sigmas : LogSigma[]
Returns: int<MeasureProduct<batch, MeasureOne>> * (float<MeasureProduct<-logL, MeasureOne>> * Point) list * LogSigma[]
|
Adaptive-metropolis-within-Gibbs algorithm, which can work in both adaptive and fixed modes. Adaptive mode: computes per‑parameter acceptance rates and tunes sigmas accordingly, repeating until all acceptance rates are in the target range. Non‑adaptive mode: checks for linear trends in parameter values over batches (via regression p‑values) and repeats until no significant trends remain.
|