Simplify API or add narrative explanation

[richelbilderbeek]

Feedback dwinter:

Both the user-interface and the code for specifying distributions is very complex, and I am concerned about how easily this package will be to maintain. @bjoelle has made some suggestions about a more modular approach to the internal code and possibly grouping like-functions into files. I think this would be a good idea, I also wonder if you could also consider a more high-level approach to specifying these models in user-exposed code. For instance, how much power/flexibility would be lost by losing the create_*_param() functions and instead doing something like this:

node_calibration <- create_gamma_distr(alpha=1, beta=10, estimate=FALSE)

I appreciate there may be reasons for taking the approach you have, but the current does make for a complex API. Whatever the API, some narrative examples
of setting up these paramaters in the documentation would be helpful.

[richelbilderbeek]

This Issue is the reason for

• creating lumier
• creating the babette YouTube channel

[richelbilderbeek]

I agree with dwinter to simplify the interface. His suggestion is:

create_gamma_distr(alpha = 1, beta = 10, estimate = FALSE)

Taking a look at BEAUti shows this is an oversimplifictation, because both/either/neither of alpha and beta can be estimated:

An alternative interface to arrive at would then be:

create_gamma_distr(alpha = 1, beta = 10, estimate_alpha = FALSE, estimate_beta = FALSE)

But this is not a great improvement over what is implemented currently:

create_gamma_distr(
  alpha = create_alpha_param(value = 1, estimate = FALSE),
  beta = create_beta_param(value = 10, estimate = FALSE)
)

[richelbilderbeek]

Perhaps a different improvement is to rename alpha to alpha_param (and for other parameters as well). From that appoach, a user can conclude what to do...

[richelbilderbeek]

Thanks to #45 ('Hyper parameters are not estimated') must parameters can be upped to a structure from only a value. The additional attributes are:

• ID: should be NA
• estimate: will be FALSE (thus no hyper parameter)

This will simplify the interface a lot.