[feature-request] Adapt S3 method conditional_effects to bmmfit objects.

[GidonFrischkorn]

Discussed in #200

^{Originally posted by GidonFrischkorn April 4, 2024}
The conditional_effects methods works ok for most bmmodels, but I think we could add some tweaks to integrate it more seamlessly into the workflow of bmm. For example, currently users need to know explicitly if a model parameter is specified as a dpar or nlpar when calling conditional_effects. Moreover, for dpar the conditional_effects are transformed back to the native scale using the specified links, this is not the case for nlpars.

Here are some examples, for conditional_effects plots for the two-parameter mixture model.

conditional_effects(ZL_fit, effects = "setsize", dpar = "kappa1") yields

whereas conditional_effects(ZL_fit, effects = "setsize", nlpar = "kappa") yields:

For theta1 versus thetat i thought that having the plot on the probability scale is super useful, but again that only works when calling it via the dpar: conditional_effects(ZL_fit, effects = "setsize", dpar = "theta1")

when called using the nlpar: conditional_effects(ZL_fit, effects = "setsize", nlpar = "thetat")

Ideally, we have an implementation of conditional_effects for bmmfit objects that uses a par argument to specify for which model parameter the effects should be plotted and maybe add an additional scale argument to specify if the plots should be given on the parameter/sampling scale or the native scale.

What do you think @venpopov?

The goal is to have a conditional_effects.bmmfit method that eases the extractions of conditional effects from bmmfit objects.