Pass argument from fit_seromodel() to rstan::sampling() via ...

[Bisaloo]

At the moment, it seems that several arguments are passed from fit_seromodel() to rstan::sampling() directly, without any transformation.

Additionally, some rstan::sampling() are hardcoded, even though users might want to adjust them (e.g., number of chains).

The standard way to deal with this, allowing for maximum flexibility while keeping the fit_seromodel() function signature concise and clear, would be to pass argument to rstan::sampling() via ....

fit_seromodel <- function(
  serodata,
  foi_model,
  decades = 0,
) {
  fit <- rstan::sampling(
    model,
    data = stan_data,
    ...
  )
}

Additional reading on ...: https://adv-r.hadley.nz/functions.html#fun-dot-dot-dot

[ntorresd]

@Bisaloo, could you please confirm to me whether this was adequately addressed by #157?

[Bisaloo]

Partially, I would also encourage you to simplify the fit_seromodel() signature even more to become:

fit_seromodel <- function(
    serodata,
    foi_model = c("constant", "tv_normal_log", "tv_normal"),
    ...
) {

However, here are two reasons you may have to keep some argument explicit rather than folding them into ...:

• if you want to pre-process them or have different actions based on their value, such as suggested in

  serofoi/R/modelling.R

  Lines 241 to 242 in 05ad912

  	# TODO Add a warning because there are exceptions where a minimal amount of
  	# iterations is needed

• if you really want to provide defaults but that's always' a difficult design choice. For example here, I definitely would not add a default for the seed, and would probably rely on rstan::sampling() existing defaults for chains, iter, thin, adapt_delta and max_treedepth()

[ntorresd]

The current development version (see #200) already addresses this issue.

[...] I would also encourage you to simplify the fit_seromodel() signature even more to become:

fit_seromodel <- function(
    serodata,
    foi_model = c("constant", "tv_normal_log", "tv_normal"),
    ...
) {

Besides of the model specification parameters, this is basically how fit_seromodel() looks like now:

fit_seromodel <- function(
  serosurvey,
  [model specification parameters],
  ...
)

Currently, we are relying on rstan::sampling defaults, as you suggested @Bisaloo . We are trying to be very explicit about checking convergence criteria; it's the user's responsability to check whether their results are statistically meaningful or not either using tools provided by the package or other methods proper of Bayesian inference.