MCMCSAMP

Name

mcmcsamp

Purpose (one Line Only)

Markov-chain Monte-Carlo sampling tool

Description

Calling Sequence

mcmcsamp,vals,scale,region,pdf

Inputs

vals - n-dimensional vector with initial guess for solution
scale - inital estimate of size of variation to sample, roughly 2x of
the expected standard deviation. One value per input variable.
If burn-in is running, this input information is modified and will
contain the final dynamically determined values.
region - size of sampling region, roughly 5x bigger than scale. One value
per input variable.

Optional Input Parameters

Keyword Input Parameters

FUNCTION_NAME - [REQUIRED!!!] Name of function that evaluates the merit
function being minimized and sampled. This function must take
a single vector of input (ie., vals) and will return a scalar
value (typically a chi-squared metric).
NSTEPS - Number of samples to collect in the chain. Default=1000
NOBURNIN - Flag, if set, will suppress any burn-in calculations and
vals/scale are assumed to already be fully tuned and ready
to go. With judicious use of the input arguments you can,
if desired, fully control the burn-in process if the built-in
algorithm is not effective.
VERBOSE - Flag, if set, will generate some verbose comments to be printed
during operation. Default is no printed output.
DISPLAY - Flag, if set, will provide output graphics to watch the sampling
process.

Outputs

pdf - [NxM] array that is the Markov chain. N is the dimenison of vals
and M is the number of steps requested.

Keyword Output Parameters

ACCEPTANCE - Scalar value that is the total acceptance rate for the returned
chain (does not include burn-in).

Common Blocks

Side Effects

Restrictions

Procedure

Modification History

2016/12/22, Written by Marc W. Buie, Southwest Research Institute with
input from Alex Parker.
2017/01/06, modified to allow imposing restricted domains for sampling