two sets of shape hyperparameters for the Beta hyperprior in the SBM

[sekulovskin]

I have implemented the two sets of shape hyperparameters. @MaartenMarsman me know what you think. If this approach is the way to go, then I will also include the new arguments in the roxygen of bgms.R.

[MaartenMarsman]

Thanks for this update @sekulovskin. I have three suggestions for the code update, see my comments:

1. The null check on between parameters I think always stops bgm because between parameters default to null. I think it is better to not pass null values (see comment 3).
2. There is a change in NUTS default, perhaps this is due to a recent update to bgms main. Could you revert this?
3. I suggest to not add a switch to all SBM functions, but instead simply pass between parameters everywhere. Then full control over the model is in the R interface, and the c++ code update simplifies. Let me know if you need help with this!

~ Maarten

[MaartenMarsman]

If I read this correctly, this will stop the analysis if the between parameters are left at their defaults. This means users cannot use bgms, except when they set other values for these defaults, irrespective if they use an SBM prior or not.

[sekulovskin]

No, this is in case the user specifies only one of the between parameters, but not both. Otherwise running the code below did work. But it doesn't matter now, since as you requested, both sets of hyperparameters are always provided.

fit <- bgm(Wenchuan[c(1:100), c(1:5)], edge_prior = "Stochastic-Block",
beta_bernoulli_alpha = 1,
beta_bernoulli_beta = 1,
beta_bernoulli_alpha_between = NULL,
beta_bernoulli_beta_between = NULL,
update_method = "adaptive-metropolis",
iter = 1000)