D function returning zero

Asked May 13 '19 at 13:38

Active May 13 '19 at 13:38

Viewed 33 times

I am attempting to differentiate the log likelihood function of a multivariate normal distribution,

NLLmvnorm <- function(data, mu, sigma) {
  a <- qr(sigma)
  logdet <- sum(log(abs(diag(a$qr))))
  sigma.inv <- fast.ginv(sigma)
  -0.5 * (logdet + t(data-mu) %*% sigma.inv %*% (data-mu))
}

Using D, I differentiate with respect to mu and sigma and store each result in a separate variable:

du <- D(expression(NLLmvnorm),"mu")
ds <- D(expression(NLLmvnorm),"sigma")

However, both of these results return 0. Am I doing something incorrect, or am I misinterpreting what D does? If it helps, my goal is to find the roots of du and ds to find maximum likelihood using solve.

asked May 13 '19 at 13:38

hkj447

See the help page, `?D`, the last paragraph before section `Value`. Your function is not among the functions `D` knows. – Rui Barradas May 13 '19 at 13:53
I see, so the only way to go about this is manually calculate the derivative and then plug that formula into R? – hkj447 May 13 '19 at 13:57
Yes, it is. But there are solvers that don't need the gradient, such as `optimise`. – Rui Barradas May 13 '19 at 13:58

D function returning zero

0 Answers0