Extend the ability of hgengamma with simple change to how hgengamma is calculated

If we calculate the log scale density and survival for generalized gamma, then exponentiate, we can avoid NaN for a wider range of mu values.

library(flexsurv)
#hgengamma <- function (x, mu = 0, sigma = 1, Q) 
#{
#  dgengamma(x = x, mu = mu, sigma = sigma, Q = Q)/pgengamma(q = x, 
#                                                            mu = mu, sigma = sigma, Q = Q, #lower.tail = FALSE)
#}

hgengamma2 <- function (x, mu = 0, sigma = 1, Q) 
{
  logdens <- dgengamma(x = x, mu = mu, sigma = sigma, Q = Q, log=TRUE)
  logsurv <- pgengamma(q = x, mu = mu, sigma = sigma, Q = Q, lower.tail = FALSE, log.p=TRUE)

  loghaz <- logdens - logsurv
  haz <- exp(loghaz)
  haz
}

## we get same answer for moderate mu
hgengamma( 10, mu= -1, sigma=1.5, Q=1) 
hgengamma2(10, mu= -1, sigma=1.5, Q=1) 

## we are able to calculate answers for more extreme mu
hgengamma( 10, mu= -14, sigma=1.5, Q=1) ## gives NaN
hgengamma2(10, mu= -14, sigma=1.5, Q=1) ## gives Answer

chjackson / flexsurv

Extend the ability of hgengamma with simple change to how hgengamma is calculated #94