johnWilshire
commented
6 years ago A problem might arise in the cdf, whilst there are c++ versions of the log likelihood I have had trouble finding a c++ version of the cdf and the cdf inverse which is necessary for pit residual and pitsboot sampling
but estimating beta should not be too hard.
The tweedie family is a very flexible family of distributions, We are only interested in tweedie distributions with power between 1 and 2, this means that they have mass at 0 but otherwise are similar to a gamma distribution,
Useful Links
There is an R package tweedie that is mostly fortran
a paper that describes ML estimation of various parameters: Dunn and Smyth 2005
a paper describes the unit deviance function and the link functions Application of the Tweedie distribution to zero-catch data in CPUE analysis