DRAMRUN  Metropolis-Hastings MCMC run with adaptive delayed rejection (DRAM)
 
  This function generates MCMC chain using DRAM adaptation for a model defined
  by user supplied sum-of-squares function and with additive i.i.d. Gaussian
  errors for the observations. The error variance sigma2 is updated
  using conjugate inverse gamma distribution.
 
  [results,chain,s2chain]=dramrun(model,data,params,options)
 
  input:
 
  model.ssfun =    ; % sum-of-squares function, ss=ssfun(par,data),
                     % that returns  -2*log(p(y|par))
  model.priorfun = ; % prior "sum-of-squares", priorfun(par,params),
                     % that returns -2*log(p(par)),
                     % default: inline('0','x','params')
 
  data   = ;         % extra argument for ssfun (to pass the data etc.)
 
  params.par0   =  ; % initial parameter vector (a row vector)
  params.sigma2 =  1;% initial/prior value for the Gaussian error variance
  params.n0     = -1;% precision of sigma2 as imaginative observations
                     %   if n0<0, no sigma2 update
  params.n      = ;  % number of actual observations (for sigma2 update)
  params.bounds = ;  % 2*npar matrix of parameter bounds
                     % default: [-Inf,Inf]
 
  options.nsimu  = 2000;   % length of the chain
  options.qcov   = ;       % proposal covariance matrix
 
  parameters for DRAM
  options.adaptint = 10;  % how often to adapt, if zero, no adaptation
  options.drscale  = 3;   % scale for the second proposal, if zero, no DR
 
  output:
 
  results  structure that contains some info about the run
  chain    nsimu*npar MCMC chain
  s2chain  sigma² chain (if generated)