DESCRIPTION:

t-mcmc-inf is an MCMC simulation engine which models data Y by a t distribution with the
following parameters:

 	degrees of freedom 	= nu
 	location 		= mu
 	scale factor 		= sigma 

Specifically, it models the data as having a sampling distribution:

p(y_i | mu, sigma, nu) = Gamma((nu+1)/2) / (sigma*sqrt(pi*nu)*Gamma(nu/2)) * [ 1 + 1/nu * (y_i - mu)^2/sigma^2 ] ^ (-(nu+1)/2)

The program runs a random walk Metropolis sampler, drawing posterior samples for mu, sigma, and nu, assuming the 
following prior distributions.  We assume independence in the prior : p(mu,sigma,nu) = p(mu)p(sigma)p(nu), and

 	p(mu) = c			(uniform)
 	p(sigma) = c/sigma		(uniform on ln(sigma))
 	p(nu) = Uniform(2, A)		(A chosen to be large enough not to truncate log likelihood)

The sampler updates each parameter at a time, using a univariate Gaussian proposal distribution on the transformed 
parameter vector (mu, lambda, gamma), where:

	mu 		= mu
	lambda 		= ln(sigma)
	gamma		= ln(nu)

INPUTS:

A file with data samples Y, one sample per line.

OUTPUTS:

A file with samples for mu, sigma, and nu, in column format, suitable for
input into R.

Run "t-mcmc-inf --helpshort" for more information on the options.

QUESTIONS?

Contact: vadimv@numberjack.net.
