Open lrnv opened 3 months ago
Usage with turing could be smoothed out by the existance of a propepr bijector, e.g. in a package extension.
Basically, Copulas are distributions functions supported on the unit d-variate hypercube, so I guess a stacked logit transfrom could be enough.
But then for SklarDist, the support information should be the same as for a Distributions.product().
Distributions.product()
The API for multivariate distributions to define their supports is not very clear to me, I'd love to get a few pointers on how to do it correctly
Refs:
https://turing.ml/dev/docs/using-turing/advanced#how-to-define-a-customized-distribution https://github.com/TuringLang/Bijectors.jl
Usage with turing could be smoothed out by the existance of a propepr bijector, e.g. in a package extension.
Basically, Copulas are distributions functions supported on the unit d-variate hypercube, so I guess a stacked logit transfrom could be enough.
But then for SklarDist, the support information should be the same as for a
Distributions.product().The API for multivariate distributions to define their supports is not very clear to me, I'd love to get a few pointers on how to do it correctly
Refs:
https://turing.ml/dev/docs/using-turing/advanced#how-to-define-a-customized-distribution https://github.com/TuringLang/Bijectors.jl