ebpm_point_gamma produces warnings and poor convergence

[stephens999]

try:

set.seed(123)
x = rpois(1000,1000)
ebpm_point_gamma(x,1)
ebpm_point_gamma(x,1)$posterior$mean

[zihao12]

Thanks! It should be fixed now. It is numerical issue for computing log(1+exp(x)) where x is small.

[pcarbo]

# logpexp(x) returns log(1 + exp(x)). The computation is performed in a
# numerically stable manner. For large entries of x, log(1 + exp(x)) is
# effectively the same as x.
logpexp <- function (x) {
  y    <- x
  i    <- which(x < 16)
  y[i] <- log(1 + exp(x[i]))
  return(y)
}