Negative Binomial Likelihood implementation

[peiyaoli]

Hi, everyone,
In GPy, it seems Negative Binomial Likelihood has not been implemented, if I want to use that, should I follow the template of poisson likelihood?

Thanks

[avehtari]

Poisson is a good starting point, but since negbin has an additional parameter, you could look also one of the other non-Gaussian likelihoods with an additional parameter.

Note also that negbin is not log-concave with all parameter-observation combinations, and thus you need to be careful when using it with Laplace, variational, and EP approximations (which is same for, e.g., student_t).

[sdwfrost]

@peiyaoli did you get negative binomial likelihoods working in GPy?