Does anyone have a suggestion for how to fit a generalized extreme value
distribution in jags?
In BUGS there is a dgev() function - but I've found it to be very buggy. It
never seems to run for me.
I could simply write the pdf of the function into the model deterministically
(setting the parameters as stochastic nodes). But if I go that route, I don't
a) are there are draw backs to keep in mind?
b) how can I deal with the fact that the pdf is discontinuous (i.e. if eta =
0, then the pdf is represented differently than if it is nonzero)?
Any suggestions would be hugely appreciated.
I'm not familiar with that distribution, but I've had reasonable success
coding nonincluded distributions using the "ones trick."
Also, discontinuous or piecewise functions can be implemented using the step
function. For example, if you have a function with a domain of , and your
function f(x) has the value g(x) when x=0 and h(x) otherwise, you can code it
f<- step(-x)g(x) + step(x)h(x)
You can use more complex combinations of step functions to create more complex
piecewise probability functions.
I'll give the step function a shot and post the model if/when I get it
I know it has been a while, but did you ever get that code working? I have a problem where the generalized extreme value distribution would probably be useful.
I wanted to give a shout-out for an implementation of extreme value distributions in JAGS, possibly the dGPD and dGEV like BUGS, but I'd be happy with just an implementation of the Gumbel distribution.
I'll make a note of it.
Sign up for the SourceForge newsletter:
You seem to have CSS turned off.
Please don't fill out this field.