Getting Interpretable Values for Posterior Draws of Parameters with Link Functions

[theonlydvr]

Most of my model parameters use link functions to maintain bounds but this makes it pretty difficult to interpret values or do custom simulations. I can get average values for the transformed parameters using the Bambi "interpret.predictions" method. Is there an equivalent to get transformed values for each posterior draw instead?

[digicosmos86]

Hi @theonlydvr,

HSSM is a wrapper for Bambi, so you can use all bambi utilities. All you need is to access the bambi model through hssm_model.model. So for example, you should be able to use interpret.predictions(hssm_model.model, idata, ...).

Please let us know if this works

[theonlydvr]

Hi @digicosmos86

I was able to use interpret.predictions but that only seems to give mean estimates for each parameter at each trial but I was hoping to get corrected estimates for each parameter at every posterior draw. Do you know how I could do that with Bambi/HSSM?

[digicosmos86]

Ah sorry for misunderstanding your post. We currently don't have anything that does that in HSSM yet, but looking at interpret.predictions in bambi, have you tried specifying the pps, target, and transforms arguments when you are using the function? Although the documentation is a bit unclear, it does seem that you can apply transforms when you are calling this function, and it also seems that setting pps to True might help get estimates for each posterior draw?

[AlexanderFengler]

@theonlydvr,

check out the add_likelihood_parameters_to_idata() function.
This will add the trial wise parameters to the posterior (post link).

You can use those directly for simulation downstream.

[theonlydvr]

Perfect, I'll definitely try that out!