Standardize function arguments where possible

[mjskay]

Some packages use different names for the same argument (e.g. "nsamples" or "draws" for number of samples to draw for the prediction / fit). Should make a complete catalog of these and standardize them

List so far (checked once normalized):

• n
  • [x] nsamples in brms
  • [x] draws in rstanarm
• re_formula
  • [x] re_formula in brms
  • [x] re.form in rstanarm
  • [ ] marginal in MCMCglmm (possibly)
• scale (for fitted_samples only)
  • [x] scale in brms
  • [x] transform in posterior_linpred
  • [ ] type in MCMCglmm
• dpar
  • [x] dpar in brms
• allow_new_levels
  • [x] allow_new_levels in brms
  • [x] no equivalent in rstanarm (always allows new levels)
  • [ ] probably no equivalent in MCMCglmm
• [x] seed
  • pattern after usage in brm(): make subsampling with n = reproducible
  • low priority (could always just call set.seed anyway)

[tmastny]

I think auxpars needs to be added to this list. add_predicted_samples.R technically swaps a auxpars call for dpar. In fact, this could lead to the same issue as in https://github.com/mjskay/tidybayes/pull/96, where if the user specified dpars in ... we'd get the matched by multiple actual arguments error.

I'm working on this fix with the pull request. Having this list is very helpful.

[mjskay]

Good point, added it to the list

[mjskay]

What remains here is post-CRAN (mostly MCMCglmm stuff, which I am punting to post-CRAN), so I am removing the CRAN milestone from this for now