softmax.R

## from http://tr.im/hH5A
logsumexp <- function (x) {
  y = max(x)
  y + log(sum(exp(x - y)))
}

softmax <- function (x) {
  exp(x - logsumexp(x))
}

Why is this used instead of the more readable exp(x) / sum(exp(x))?

@tmalsburg Numerical stability. Consider

softmax2 <- function(x) exp(x) / sum(exp(x))

softmax(c(1, 2, 3) * 1000)  # NaN NaN NaN
softmax2(c(1, 2, 3) * 1000)  # 0 0 1

@ben519 maybe the opposite?
softmax2 <- function(x) exp(x) / sum(exp(x))
softmax2(c(1, 2, 3) * 1000) # NaN NaN NaN
softmax(c(1, 2, 3) * 1000) # 0 0 1

If the exponent exceeds a size of the defined variable the specified error occurs. I noticed that the number x=90 and larger leads to error for type of variable=float that holds 4B. So I'm doing the cutting to 80 and all going nice. This Sofmax code that clarifies this written in c is on link:
https://github.com/DusanRandD/Softmax