R/dMit.R

In AdMit: Adaptive Mixture of Student-t Distributions

## Function which computes the density of a mixture of Student-t densities
## __input__
## theta : [Nxk matrix] of values
## mit   : [list] of mixture information (default: univariate Cauchy)
## log    : [logical] log output (default: TRUE)
## __output__
## [Nx1 vector] of density evaluated at theta
## __20080329__
## changes marked by # date 20120816 (updated mit definition: mit$df can be a double or vector)
'dMit' <- function(theta, mit=list(), log=TRUE)
{
  if (missing(theta))
    stop ("'theta' is missing in 'dMit'")
  H <- length(mit$p)
  if (H==0) 
  { ## default for the mixture
    warning ("'mit' not well defined in 'dMit'; set to default")
    mit <- list(p=1, mu=as.matrix(0), Sigma=as.matrix(1), df=1)
    H <- 1
  }
  if (is.vector(theta))
  { ## if vector is supplied instead of a matrix
    if (ncol(mit$mu)==1)
    { ## univariate density
      theta <- as.matrix(theta)
    }
    else
    { ## multivariate density but evaluated at a single point
      theta <- matrix(theta, nrow=1)
    }
  }
  r <- tmp <- 0
  # date 20120816: df can be a vector of size 1 or H, replicated in the former case
  if(length(mit$df)==1 & H >1)
    mit$df = rep(mit$df,H)
  for (h in 1:H) {
    # date 20120816: density evaluation of mixture with possibly different df parameters
    #tmp <- exp(log(mit$p[h]) + fn.dmvt(theta, mit$mu[h, ], 
    #    mit$Sigma[h, ], mit$df, log = TRUE))
    tmp <- exp(log(mit$p[h]) + fn.dmvt(theta, mit$mu[h, ], 
                                       mit$Sigma[h, ], mit$df[h], log = TRUE))
    r <- r + tmp
  }
  if (log) 
    r <- log(r)
  r <- as.vector(r)
  return(r)
}