Could you explain if bambi supports extra predefined function with extra hyperparameter to formulate in the model? and estimate posterior for those parameters?

[wenwenzws]

Hi,

I am trying to have my own transformation functions with parameters to be estimated as posterior from bayes linear model.
For example,
def hill_core(x, k,s):

      #  transformed_hill=1 - half**s/(x**s + half**s)
        transformed_hill=x*k*s
        return transformed_hill

bmb_model=bmb.Model('gbv ~ hill_core(sem_nonbrand_click,k,s)',model_obj.df, priors_dic)

I want to have the posterior distribution of k and s as well. Is this possible in the current formulation?

[tomicapretto]

This issue has been raised in #448 as well.

The short answer is that Bambi supports generalized linear models now. The function you wrote above suggests a non-linear model, which is still not supported in Bambi.

We do plan to support it at some point in time, but I'm afraid it's not coming in the short term (few months from now).
See #544 for some ideas we have.