David Widmann
David Widmann
IMO it is a bit problematic that the method requires a parameter `:lp` with the log-likelihood. The name is not standardized, i.e., it might not exist or it might denote...
I totally agree that it would be good to make the statistical methods more modular - this was also the main motivation for changing (most) implementations in MCMCChains to operate...
Yep, I think that sounds good. I would just suggest ```julia function dic(chain::Chains, loglik::Symbol) lps = Array(chain[:, loglik, :]) return dic(lps) end ``` instead.
> MCMCChains tests on Julia 1.0.5, but StanDump requires Julia 1.1 or higher. What is the best path forward? Do you know why StanDump requires Julia 1.1?
I personally don't need 1.0 (or anything apart from the latest version) but I think it would be unfortunate to drop Julia 1.0 if it can be supported without much...
> The current version of StatsModelComparisons.jl uses StanSample.jl for testing purposes. I.e., it is used only in the tests? Then it should not be listed among the dependencies but in...
Yes, since the packages listed under `extras` are not pulled in and installed when the package is installed - they are only used for testing.
I opened a PR that also uses RecipeBase instead of StatsPlots: https://github.com/StatisticalRethinkingJulia/StatsModelComparisons.jl/pull/4
I noticed that in the mean time the license of StatsModelComparisons was changed to GPL-3 - does this mean we would have to relicense `MCMCChains.dic` (or maybe even other parts?)...
@cpfiffer What's your opinion about the license issue?